From e6841db50b7a26f45d69c53987a9bd06feb4936b Mon Sep 17 00:00:00 2001
From: Hakula Chen <i@hakula.xyz>
Date: Fri, 29 May 2026 16:50:37 +0800
Subject: [PATCH 1/5] feat(model): add Claude Opus 4.8 as the default model

Opus 4.8 shares 4.7's full capability set (interleaved thinking, context
management, 1M context, the Low..Max effort ladder, structured outputs) and
the same $5/$25 pricing, so a single most-specific-first MODELS row makes it
fully functional: betas, pricing, context window, and the auto-compaction
threshold all derive from the table.

Make it the default model and the `opus` alias target. The combined
model+effort picker lists 4.8 in place of 4.7 (4.7 stays selectable via the
table), mirroring how 4.6 was dropped from the roster when 4.7 shipped.

oxide-code keeps its highest-non-Max effort heuristic, so 4.8 defaults to
xhigh rather than upstream's high default.
---
 .../oxide-code/src/client/anthropic/betas.rs  | 13 ++++
 crates/oxide-code/src/config.rs               |  5 +-
 crates/oxide-code/src/model.rs                | 60 +++++++++++++------
 crates/oxide-code/src/prompt/environment.rs   |  1 +
 crates/oxide-code/src/slash.rs                |  2 +-
 crates/oxide-code/src/slash/model.rs          | 32 +++++-----
 crates/oxide-code/src/slash/picker.rs         | 20 +++----
 ...er_cancelling_shows_spinner_and_label.snap |  2 +-
 ...acting_shows_spinner_and_status_label.snap |  2 +-
 ...med_shows_static_hint_without_spinner.snap |  2 +-
 ..._with_title_shows_model_title_and_cwd.snap |  2 +-
 ...idle_without_title_leaves_slot_unused.snap |  2 +-
 ...w_width_preserves_model_and_run_state.snap |  2 +-
 ...eaming_shows_spinner_and_status_label.snap |  2 +-
 ...us__tests__render_tool_running_status.snap |  2 +-
 ...m_full_layout_still_includes_starters.snap |  2 +-
 ...ollapsed_drops_box_but_keeps_starters.snap |  2 +-
 ..._box_environment_starters_and_trailer.snap |  2 +-
 ...rters_and_truncates_cwd_in_the_middle.snap |  2 +-
 .../oxide-code/src/tui/components/status.rs   | 18 +++---
 .../oxide-code/src/tui/components/welcome.rs  |  4 +-
 docs/design/tui/status-line.md                |  2 +-
 docs/design/tui/welcome.md                    |  4 +-
 docs/guide/configuration.md                   | 16 ++---
 docs/research/api/anthropic.md                |  6 +-
 docs/research/tui/status-line.md              | 12 ++--
 26 files changed, 130 insertions(+), 89 deletions(-)

diff --git a/crates/oxide-code/src/client/anthropic/betas.rs b/crates/oxide-code/src/client/anthropic/betas.rs
index e9ce667a..79e21ae6 100644
--- a/crates/oxide-code/src/client/anthropic/betas.rs
+++ b/crates/oxide-code/src/client/anthropic/betas.rs
@@ -209,6 +209,19 @@ mod tests {
         assert!(with_1m.contains(&CONTEXT_1M_BETA_HEADER));
     }
 
+    #[test]
+    fn compute_betas_opus_4_8_matches_opus_4_7_header_order() {
+        // 4.8 shares 4.7's capability flags, so the fingerprinted header sequence must be
+        // byte-identical (3P gateways reject mismatches).
+        assert_eq!(
+            compute_betas("claude-opus-4-8", &api_key(), true, false),
+            compute_betas("claude-opus-4-7", &api_key(), true, false),
+        );
+        let with_1m = compute_betas("claude-opus-4-8[1m]", &api_key(), true, false);
+        assert!(with_1m.contains(&CONTEXT_1M_BETA_HEADER));
+        assert!(with_1m.contains(&EFFORT_BETA_HEADER));
+    }
+
     #[test]
     fn compute_betas_structured_outputs_gated_by_model_capability() {
         assert_eq!(
diff --git a/crates/oxide-code/src/config.rs b/crates/oxide-code/src/config.rs
index 627b10bb..df31713c 100644
--- a/crates/oxide-code/src/config.rs
+++ b/crates/oxide-code/src/config.rs
@@ -17,7 +17,7 @@ use crate::tui::theme::{self, Theme};
 use crate::util::env;
 use crate::util::path::expand_user;
 
-const DEFAULT_MODEL: &str = "claude-opus-4-7[1m]";
+const DEFAULT_MODEL: &str = "claude-opus-4-8[1m]";
 const DEFAULT_BASE_URL: &str = "https://api.anthropic.com";
 const AUTO_COMPACTION_OUTPUT_RESERVE_CAP: u32 = 20_000;
 const AUTO_COMPACTION_BUFFER_TOKENS: u32 = 13_000;
@@ -879,7 +879,7 @@ mod tests {
 
     #[tokio::test]
     async fn load_defaults_apply_when_no_config_and_no_env() {
-        // Opus 4.7 supports `xhigh`; `effort` / `max_tokens` derive from that ceiling.
+        // Opus 4.8 supports `xhigh`; `effort` / `max_tokens` derive from that ceiling.
         let dir = tempfile::tempdir().unwrap();
         let config = temp_env::async_with_vars(env_vars(vec![xdg(&dir)]), Config::load())
             .await
@@ -1507,6 +1507,7 @@ mod tests {
     #[tokio::test]
     async fn load_effort_default_follows_model_ceiling() {
         for (model, expected) in [
+            ("claude-opus-4-8", Some(Effort::Xhigh)),
             ("claude-opus-4-7", Some(Effort::Xhigh)),
             ("claude-opus-4-6", Some(Effort::High)),
             ("claude-sonnet-4-6", Some(Effort::High)),
diff --git a/crates/oxide-code/src/model.rs b/crates/oxide-code/src/model.rs
index f17ef0a1..7abd6a9c 100644
--- a/crates/oxide-code/src/model.rs
+++ b/crates/oxide-code/src/model.rs
@@ -51,6 +51,25 @@ pub(crate) struct Capabilities {
 
 /// Most-specific substring first.
 pub(crate) const MODELS: &[ModelInfo] = &[
+    ModelInfo {
+        id_substr: "claude-opus-4-8",
+        display_name: "Claude Opus 4.8",
+        cutoff: Some("January 2026"),
+        capabilities: Capabilities {
+            interleaved_thinking: true,
+            context_management: true,
+            context_1m: true,
+            supported_efforts: &[
+                Effort::Low,
+                Effort::Medium,
+                Effort::High,
+                Effort::Xhigh,
+                Effort::Max,
+            ],
+            structured_outputs: true,
+        },
+        cost_rates: Some(OPUS_4_5_PLUS_RATES),
+    },
     ModelInfo {
         id_substr: "claude-opus-4-7",
         display_name: "Claude Opus 4.7",
@@ -279,10 +298,10 @@ mod tests {
 
     #[test]
     fn capability_flags_match_upstream_substring_predicates() {
-        // Locks substring-derived flags to upstream's `modelSupports*` predicates. Opus 4.7
-        // postdates the predicate set and is skipped.
+        // Locks substring-derived flags to upstream's `modelSupports*` predicates. Opus 4.7 and
+        // 4.8 postdate the predicate set and are skipped.
         for info in MODELS {
-            if info.id_substr == "claude-opus-4-7" {
+            if matches!(info.id_substr, "claude-opus-4-7" | "claude-opus-4-8") {
                 continue;
             }
             let m = info.id_substr;
@@ -306,15 +325,17 @@ mod tests {
     }
 
     #[test]
-    fn opus_4_7_uniquely_supports_xhigh() {
-        // Upstream predates 4.7; pin so a future "alignment" edit doesn't strip our caps.
-        let caps = lookup("claude-opus-4-7").unwrap().capabilities;
-        assert!(caps.interleaved_thinking);
-        assert!(caps.context_management);
-        assert!(caps.context_1m);
-        assert!(caps.accepts_effort(Effort::Xhigh));
-        assert!(caps.accepts_effort(Effort::Max));
-        assert!(caps.structured_outputs);
+    fn opus_4_7_and_4_8_support_the_full_effort_ladder() {
+        // Upstream predates 4.7 / 4.8; pin so a future "alignment" edit doesn't strip our caps.
+        for id in ["claude-opus-4-7", "claude-opus-4-8"] {
+            let caps = lookup(id).unwrap().capabilities;
+            assert!(caps.interleaved_thinking, "{id}");
+            assert!(caps.context_management, "{id}");
+            assert!(caps.context_1m, "{id}");
+            assert!(caps.accepts_effort(Effort::Xhigh), "{id}");
+            assert!(caps.accepts_effort(Effort::Max), "{id}");
+            assert!(caps.structured_outputs, "{id}");
+        }
 
         for other in [
             "claude-opus-4-6",
@@ -329,14 +350,14 @@ mod tests {
                     .unwrap()
                     .capabilities
                     .accepts_effort(Effort::Xhigh),
-                "{other} must not accept Xhigh — it 400s on non-4.7",
+                "{other} must not accept Xhigh — it 400s on pre-4.7 models",
             );
         }
     }
 
     #[test]
     fn effort_max_is_opus_only() {
-        for supported in ["claude-opus-4-7", "claude-opus-4-6"] {
+        for supported in ["claude-opus-4-8", "claude-opus-4-7", "claude-opus-4-6"] {
             assert!(
                 lookup(supported)
                     .unwrap()
@@ -378,6 +399,7 @@ mod tests {
     #[test]
     fn structured_outputs_flag_tracks_upstream_allowlist() {
         for supported in [
+            "claude-opus-4-8",
             "claude-opus-4-7",
             "claude-opus-4-6",
             "claude-opus-4-5",
@@ -456,9 +478,11 @@ mod tests {
 
     #[test]
     fn default_effort_picks_highest_supported_tier_when_user_has_no_pick() {
-        // Opus 4.7: full ladder → xhigh.
-        let opus_4_7 = lookup("claude-opus-4-7").unwrap().capabilities;
-        assert_eq!(opus_4_7.default_effort(), Some(Effort::Xhigh));
+        // Opus 4.7 / 4.8: full ladder → xhigh.
+        for id in ["claude-opus-4-7", "claude-opus-4-8"] {
+            let caps = lookup(id).unwrap().capabilities;
+            assert_eq!(caps.default_effort(), Some(Effort::Xhigh), "{id}");
+        }
 
         // Opus 4.6 / Sonnet 4.6: effort but no xhigh → high.
         for id in ["claude-opus-4-6", "claude-sonnet-4-6"] {
@@ -613,6 +637,7 @@ mod tests {
     #[test]
     fn display_name_known_plain_id_renders_row_label() {
         for (id, expected) in [
+            ("claude-opus-4-8", "Claude Opus 4.8"),
             ("claude-opus-4-7", "Claude Opus 4.7"),
             ("claude-opus-4-6", "Claude Opus 4.6"),
             ("claude-sonnet-4-6", "Claude Sonnet 4.6"),
@@ -649,6 +674,7 @@ mod tests {
     #[test]
     fn short_display_name_strips_claude_family_prefix() {
         for (id, expected) in [
+            ("claude-opus-4-8", "Opus 4.8"),
             ("claude-opus-4-7", "Opus 4.7"),
             ("claude-sonnet-4-6", "Sonnet 4.6"),
             ("claude-haiku-4-5", "Haiku 4.5"),
diff --git a/crates/oxide-code/src/prompt/environment.rs b/crates/oxide-code/src/prompt/environment.rs
index 8ec67576..1a850be2 100644
--- a/crates/oxide-code/src/prompt/environment.rs
+++ b/crates/oxide-code/src/prompt/environment.rs
@@ -253,6 +253,7 @@ mod tests {
     #[test]
     fn knowledge_cutoff_known_models() {
         for (id, expected) in [
+            ("claude-opus-4-8", "January 2026"),
             ("claude-opus-4-7", "January 2026"),
             ("claude-sonnet-4-6", "August 2025"),
             ("claude-opus-4-6", "May 2025"),
diff --git a/crates/oxide-code/src/slash.rs b/crates/oxide-code/src/slash.rs
index aaf0f075..2b10f462 100644
--- a/crates/oxide-code/src/slash.rs
+++ b/crates/oxide-code/src/slash.rs
@@ -135,7 +135,7 @@ pub(crate) fn test_session_info() -> LiveSessionInfo {
             auth_label: "API key",
             base_url: "https://api.test.invalid".to_owned(),
             extra_ca_certs: None,
-            model_id: "claude-opus-4-7".to_owned(),
+            model_id: "claude-opus-4-8".to_owned(),
             effort: Some(Effort::High),
             max_tokens: 32_000,
             max_tool_rounds: None,
diff --git a/crates/oxide-code/src/slash/model.rs b/crates/oxide-code/src/slash/model.rs
index 2f65415f..c642724a 100644
--- a/crates/oxide-code/src/slash/model.rs
+++ b/crates/oxide-code/src/slash/model.rs
@@ -15,7 +15,7 @@ use crate::model::{MODELS, ResolvedModelId, display_name, lookup};
 const TAG_1M: &str = "[1m]";
 
 const ALIASES: &[(&str, &str)] = &[
-    ("opus", "claude-opus-4-7"),
+    ("opus", "claude-opus-4-8"),
     ("sonnet", "claude-sonnet-4-6"),
     ("haiku", "claude-haiku-4-5"),
 ];
@@ -79,7 +79,7 @@ fn resolve_model_arg(arg: &str) -> Result<ResolvedModelId, String> {
     };
     if base_arg.is_empty() {
         return Err(format!(
-            "`{TAG_1M}` is a tag, not a model. Try `/model opus{TAG_1M}` or `/model claude-opus-4-7{TAG_1M}`.",
+            "`{TAG_1M}` is a tag, not a model. Try `/model opus{TAG_1M}` or `/model claude-opus-4-8{TAG_1M}`.",
         ));
     }
     let base_id = resolve_base(base_arg)?;
@@ -238,19 +238,19 @@ mod tests {
             .into_iter()
             .map(|(v, _)| v)
             .collect();
-        assert_eq!(got, vec!["claude-opus-4-7", "claude-opus-4-7[1m]"]);
+        assert_eq!(got, vec!["claude-opus-4-8", "claude-opus-4-8[1m]"]);
     }
 
     #[test]
     fn complete_arg_appends_1m_context_suffix_only_for_1m_variants() {
-        let rows = arg_rows("claude-opus-4-7");
+        let rows = arg_rows("claude-opus-4-8");
         let one_m = rows
             .iter()
-            .find(|(v, _)| v == "claude-opus-4-7[1m]")
+            .find(|(v, _)| v == "claude-opus-4-8[1m]")
             .expect("1M variant present");
         let plain = rows
             .iter()
-            .find(|(v, _)| v == "claude-opus-4-7")
+            .find(|(v, _)| v == "claude-opus-4-8")
             .expect("plain variant present");
         assert!(
             one_m.1.contains("1M context"),
@@ -296,8 +296,8 @@ mod tests {
     #[test]
     fn execute_with_alias_resolves_to_canonical_id() {
         for (alias, expected) in [
-            ("opus", "claude-opus-4-7"),
-            ("opus[1m]", "claude-opus-4-7[1m]"),
+            ("opus", "claude-opus-4-8"),
+            ("opus[1m]", "claude-opus-4-8[1m]"),
             ("sonnet", "claude-sonnet-4-6"),
             ("sonnet[1m]", "claude-sonnet-4-6[1m]"),
             ("haiku", "claude-haiku-4-5"),
@@ -391,7 +391,7 @@ mod tests {
             );
         }
         assert!(
-            msg.contains("Claude Opus 4.7 (1M context)"),
+            msg.contains("Claude Opus 4.8 (1M context)"),
             "1M variant renders the (1M context) suffix: {msg}",
         );
         assert_eq!(chat.entry_count(), 0);
@@ -427,8 +427,8 @@ mod tests {
     fn execute_ambiguous_listing_falls_back_to_full_curated_set_when_filter_empty() {
         let (_, outcome) = run_execute("claude-opus");
         let msg = outcome.expect_err("ambiguous arg must error");
-        // `claude-opus` matches the listed Opus 4.7 entries, so the listing surfaces those.
-        for id in ["claude-opus-4-7", "claude-opus-4-7[1m]"] {
+        // `claude-opus` matches the listed Opus 4.8 entries, so the listing surfaces those.
+        for id in ["claude-opus-4-8", "claude-opus-4-8[1m]"] {
             assert!(msg.contains(id), "{id} should be listed: {msg}");
         }
         // Older non-listed Opus rows must not appear in the curated listing.
@@ -447,7 +447,7 @@ mod tests {
             resolve_model_arg("opus")
                 .as_ref()
                 .map(ResolvedModelId::as_str),
-            Ok("claude-opus-4-7")
+            Ok("claude-opus-4-8")
         );
     }
 
@@ -489,7 +489,7 @@ mod tests {
             resolve_model_arg("OPUS")
                 .as_ref()
                 .map(ResolvedModelId::as_str),
-            Ok("claude-opus-4-7")
+            Ok("claude-opus-4-8")
         );
         assert_eq!(
             resolve_model_arg("Claude-Opus-4-7")
@@ -501,7 +501,7 @@ mod tests {
             resolve_model_arg("OPUS[1M]")
                 .as_ref()
                 .map(ResolvedModelId::as_str),
-            Ok("claude-opus-4-7[1m]"),
+            Ok("claude-opus-4-8[1m]"),
         );
     }
 
@@ -543,11 +543,11 @@ mod tests {
     fn listed_models_matching_surfaces_1m_variants_alongside_base() {
         let out = listed_models_matching("claude-opus");
         assert!(
-            out.contains("- `claude-opus-4-7` — Claude Opus 4.7"),
+            out.contains("- `claude-opus-4-8` — Claude Opus 4.8"),
             "{out}"
         );
         assert!(
-            out.contains("- `claude-opus-4-7[1m]` — Claude Opus 4.7 (1M context)"),
+            out.contains("- `claude-opus-4-8[1m]` — Claude Opus 4.8 (1M context)"),
             "{out}",
         );
     }
diff --git a/crates/oxide-code/src/slash/picker.rs b/crates/oxide-code/src/slash/picker.rs
index 677595fe..79077827 100644
--- a/crates/oxide-code/src/slash/picker.rs
+++ b/crates/oxide-code/src/slash/picker.rs
@@ -22,8 +22,8 @@ use crate::tui::theme::Theme;
 /// Curated roster shown in the picker AND in `/model`'s typed-arg autocomplete; `/model <id>`
 /// still resolves against the full `MODELS` table.
 pub(super) const LISTED_MODELS: &[&str] = &[
-    "claude-opus-4-7",
-    "claude-opus-4-7[1m]",
+    "claude-opus-4-8",
+    "claude-opus-4-8[1m]",
     "claude-sonnet-4-6",
     "claude-sonnet-4-6[1m]",
     "claude-haiku-4-5",
@@ -329,10 +329,10 @@ mod tests {
 
     #[test]
     fn new_positions_cursor_on_active_model() {
-        // `test_session_info` ships claude-opus-4-7 active.
+        // `test_session_info` ships claude-opus-4-8 active.
         let p = picker();
         let row = p.list.selected().expect("active row");
-        assert_eq!(row.id, "claude-opus-4-7");
+        assert_eq!(row.id, "claude-opus-4-8");
         assert!(row.is_active);
     }
 
@@ -366,7 +366,7 @@ mod tests {
 
     #[test]
     fn right_arrow_cycles_effort_within_supported_levels() {
-        // Opus 4.7 supports the full ladder. Pressing Right walks
+        // Opus 4.8 supports the full ladder. Pressing Right walks
         // through it; Left walks back.
         let mut p = picker();
         let initial = p.effort;
@@ -396,7 +396,7 @@ mod tests {
         // — pin it independently. Cycle Left until the effort returns to the initial pick.
         let mut p = picker();
         p.handle_key(&key(KeyCode::Right)); // arm the axis with a known starting tier
-        let initial = p.effort.expect("Opus 4.7 has an effort axis");
+        let initial = p.effort.expect("Opus 4.8 has an effort axis");
         for _ in 0..16 {
             p.handle_key(&key(KeyCode::Left));
             if p.effort == Some(initial) {
@@ -439,7 +439,7 @@ mod tests {
             ModalKey::Submitted(ModalAction::User(UserAction::SwapConfig { model, effort })) => {
                 assert_eq!(
                     model.map(ResolvedModelId::into_inner).as_deref(),
-                    Some("claude-opus-4-7[1m]"),
+                    Some("claude-opus-4-8[1m]"),
                 );
                 assert!(
                     effort.is_none(),
@@ -452,7 +452,7 @@ mod tests {
 
     #[test]
     fn enter_after_effort_change_emits_swap_with_effort_only() {
-        // Opus 4.7 active + High; cycle effort Forward once → Xhigh.
+        // Opus 4.8 active + High; cycle effort Forward once → Xhigh.
         let mut p = picker();
         p.handle_key(&key(KeyCode::Right));
         let outcome = p.handle_key(&key(KeyCode::Enter));
@@ -491,10 +491,10 @@ mod tests {
     #[test]
     fn render_runs_at_typical_widths_without_panicking() {
         let theme = Theme::default();
-        // Two cursor positions: an effort-tier model (Opus 4.7) so the effort row renders, and
+        // Two cursor positions: an effort-tier model (Opus 4.8) so the effort row renders, and
         // a no-tier model (Haiku 4.5) so the hide branch executes.
         for setup in [
-            None,                     // Opus 4.7 — has effort tier
+            None,                     // Opus 4.8 — has effort tier
             Some(KeyCode::Char('5')), // Haiku 4.5 — no effort tier
         ] {
             let mut p = picker();
diff --git a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_cancelling_shows_spinner_and_label.snap b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_cancelling_shows_spinner_and_label.snap
index e2dfaa7c..ad60468c 100644
--- a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_cancelling_shows_spinner_and_label.snap
+++ b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_cancelling_shows_spinner_and_label.snap
@@ -2,5 +2,5 @@
 source: crates/oxide-code/src/tui/components/status.rs
 expression: "render_status(&mut bar, 80)"
 ---
-"  ~/projects/demo │ main │ Opus 4.7 (xhigh) │ ⣷ Cancelling                      "
+"  ~/projects/demo │ main │ Opus 4.8 (xhigh) │ ⣷ Cancelling                      "
 "────────────────────────────────────────────────────────────────────────────────"
diff --git a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_compacting_shows_spinner_and_status_label.snap b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_compacting_shows_spinner_and_status_label.snap
index 5df03297..b082e1e7 100644
--- a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_compacting_shows_spinner_and_status_label.snap
+++ b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_compacting_shows_spinner_and_status_label.snap
@@ -2,5 +2,5 @@
 source: crates/oxide-code/src/tui/components/status.rs
 expression: "render_status(&mut bar, 80)"
 ---
-"  ~/projects/demo │ main │ Opus 4.7 (xhigh) │ ⣷ Compacting · Esc to interrupt   "
+"  ~/projects/demo │ main │ Opus 4.8 (xhigh) │ ⣷ Compacting · Esc to interrupt   "
 "────────────────────────────────────────────────────────────────────────────────"
diff --git a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_exit_armed_shows_static_hint_without_spinner.snap b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_exit_armed_shows_static_hint_without_spinner.snap
index 9585ff7d..18cf0afb 100644
--- a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_exit_armed_shows_static_hint_without_spinner.snap
+++ b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_exit_armed_shows_static_hint_without_spinner.snap
@@ -2,5 +2,5 @@
 source: crates/oxide-code/src/tui/components/status.rs
 expression: "render_status(&mut bar, 80)"
 ---
-"  ~/projects/demo │ main │ Opus 4.7 (xhigh) │ Press Ctrl+C again to exit        "
+"  ~/projects/demo │ main │ Opus 4.8 (xhigh) │ Press Ctrl+C again to exit        "
 "────────────────────────────────────────────────────────────────────────────────"
diff --git a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_idle_with_title_shows_model_title_and_cwd.snap b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_idle_with_title_shows_model_title_and_cwd.snap
index a29427c9..0e179fcb 100644
--- a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_idle_with_title_shows_model_title_and_cwd.snap
+++ b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_idle_with_title_shows_model_title_and_cwd.snap
@@ -2,5 +2,5 @@
 source: crates/oxide-code/src/tui/components/status.rs
 expression: "render_status(&mut bar, 80)"
 ---
-"  ~/projects/demo │ main │ Opus 4.7 (xhigh) │ Ready │ Fix login flow            "
+"  ~/projects/demo │ main │ Opus 4.8 (xhigh) │ Ready │ Fix login flow            "
 "────────────────────────────────────────────────────────────────────────────────"
diff --git a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_idle_without_title_leaves_slot_unused.snap b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_idle_without_title_leaves_slot_unused.snap
index 9f86bb35..585b5735 100644
--- a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_idle_without_title_leaves_slot_unused.snap
+++ b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_idle_without_title_leaves_slot_unused.snap
@@ -2,5 +2,5 @@
 source: crates/oxide-code/src/tui/components/status.rs
 expression: "render_status(&mut bar, 80)"
 ---
-"  ~/projects/demo │ main │ Opus 4.7 (xhigh) │ Ready                             "
+"  ~/projects/demo │ main │ Opus 4.8 (xhigh) │ Ready                             "
 "────────────────────────────────────────────────────────────────────────────────"
diff --git a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_narrow_width_preserves_model_and_run_state.snap b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_narrow_width_preserves_model_and_run_state.snap
index 1af8661e..e28cc371 100644
--- a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_narrow_width_preserves_model_and_run_state.snap
+++ b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_narrow_width_preserves_model_and_run_state.snap
@@ -2,5 +2,5 @@
 source: crates/oxide-code/src/tui/components/status.rs
 expression: "render_status(&mut bar, 40)"
 ---
-"  Opus 4.7 (xhigh) │ Ready              "
+"  Opus 4.8 (xhigh) │ Ready              "
 "────────────────────────────────────────"
diff --git a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_streaming_shows_spinner_and_status_label.snap b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_streaming_shows_spinner_and_status_label.snap
index a522af7b..a14e2325 100644
--- a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_streaming_shows_spinner_and_status_label.snap
+++ b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_streaming_shows_spinner_and_status_label.snap
@@ -2,5 +2,5 @@
 source: crates/oxide-code/src/tui/components/status.rs
 expression: "render_status(&mut bar, 80)"
 ---
-"  ~/projects/demo │ main │ Opus 4.7 (xhigh) │ ⣷ Streaming · Esc to interrupt    "
+"  ~/projects/demo │ main │ Opus 4.8 (xhigh) │ ⣷ Streaming · Esc to interrupt    "
 "────────────────────────────────────────────────────────────────────────────────"
diff --git a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_tool_running_status.snap b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_tool_running_status.snap
index 5e260f71..c5afa39f 100644
--- a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_tool_running_status.snap
+++ b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__status__tests__render_tool_running_status.snap
@@ -2,5 +2,5 @@
 source: crates/oxide-code/src/tui/components/status.rs
 expression: "render_status(&mut bar, 80)"
 ---
-"  ~/projects/demo │ main │ Opus 4.7 (xhigh) │ ⣷ Running bash · Esc to interrupt "
+"  ~/projects/demo │ main │ Opus 4.8 (xhigh) │ ⣷ Running bash · Esc to interrupt "
 "────────────────────────────────────────────────────────────────────────────────"
diff --git a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__welcome__tests__paint_60_col_minimum_full_layout_still_includes_starters.snap b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__welcome__tests__paint_60_col_minimum_full_layout_still_includes_starters.snap
index 0635473f..765060fa 100644
--- a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__welcome__tests__paint_60_col_minimum_full_layout_still_includes_starters.snap
+++ b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__welcome__tests__paint_60_col_minimum_full_layout_still_includes_starters.snap
@@ -5,7 +5,7 @@ expression: "render(60, 14, &snap)"
 "                                                            "
 "                 ━━━━ oxide-code v0.1.0 ━━━━                "
 "                                                            "
-"           Claude Opus 4.7 · xhigh effort · OAuth           "
+"           Claude Opus 4.8 · xhigh effort · OAuth           "
 "           ~/github/oxide-code                              "
 "                                                            "
 "           Try one of:                                      "
diff --git a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__welcome__tests__paint_collapsed_drops_box_but_keeps_starters.snap b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__welcome__tests__paint_collapsed_drops_box_but_keeps_starters.snap
index 541c1e4f..85df338b 100644
--- a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__welcome__tests__paint_collapsed_drops_box_but_keeps_starters.snap
+++ b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__welcome__tests__paint_collapsed_drops_box_but_keeps_starters.snap
@@ -5,7 +5,7 @@ expression: "render(50, 14, &snap)"
 "                                                  "
 "                 oxide-code v0.1.0                "
 "                                                  "
-"      Claude Opus 4.7 · xhigh effort              "
+"      Claude Opus 4.8 · xhigh effort              "
 "      ~/github/oxide-code                         "
 "                                                  "
 "      Try one of:                                 "
diff --git a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__welcome__tests__paint_full_width_renders_box_environment_starters_and_trailer.snap b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__welcome__tests__paint_full_width_renders_box_environment_starters_and_trailer.snap
index 0c65a5e8..f46b26eb 100644
--- a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__welcome__tests__paint_full_width_renders_box_environment_starters_and_trailer.snap
+++ b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__welcome__tests__paint_full_width_renders_box_environment_starters_and_trailer.snap
@@ -5,7 +5,7 @@ expression: "render(80, 14, &snap)"
 "                                                                                "
 "                           ━━━━ oxide-code v0.1.0 ━━━━                          "
 "                                                                                "
-"                     Claude Opus 4.7 · xhigh effort · OAuth                     "
+"                     Claude Opus 4.8 · xhigh effort · OAuth                     "
 "                     ~/github/oxide-code                                        "
 "                                                                                "
 "                     Try one of:                                                "
diff --git a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__welcome__tests__paint_narrow_drops_starters_and_truncates_cwd_in_the_middle.snap b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__welcome__tests__paint_narrow_drops_starters_and_truncates_cwd_in_the_middle.snap
index 29a64cf9..a61a3aa3 100644
--- a/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__welcome__tests__paint_narrow_drops_starters_and_truncates_cwd_in_the_middle.snap
+++ b/crates/oxide-code/src/tui/components/snapshots/ox__tui__components__welcome__tests__paint_narrow_drops_starters_and_truncates_cwd_in_the_middle.snap
@@ -5,7 +5,7 @@ expression: "render(30, 8, &snap)"
 "                              "
 "       oxide-code v0.1.0      "
 "                              "
-"Claude Opus 4.7 · xhigh       "
+"Claude Opus 4.8 · xhigh       "
 "~/very/long/w...ath/oxide-code"
 "                              "
 "                              "
diff --git a/crates/oxide-code/src/tui/components/status.rs b/crates/oxide-code/src/tui/components/status.rs
index 710ee7cb..13233029 100644
--- a/crates/oxide-code/src/tui/components/status.rs
+++ b/crates/oxide-code/src/tui/components/status.rs
@@ -363,7 +363,7 @@ mod tests {
         StatusBar::new(
             &Theme::default(),
             vec![StatusLineSegment::PullRequest, StatusLineSegment::RunState],
-            "Opus 4.7".into(),
+            "Opus 4.8".into(),
             None,
             String::new(),
             None,
@@ -407,11 +407,11 @@ mod tests {
     #[test]
     fn set_model_replaces_displayed_model_label() {
         let mut bar = test_bar();
-        bar.set_model("Opus 4.7".to_owned());
-        assert_eq!(bar.model(), "Opus 4.7");
+        bar.set_model("Opus 4.8".to_owned());
+        assert_eq!(bar.model(), "Opus 4.8");
         let output = render_top_row(&mut bar, 80);
         assert!(
-            output.contains("Opus 4.7"),
+            output.contains("Opus 4.8"),
             "new label must reach the rendered bar: {output:?}",
         );
         assert!(
@@ -727,7 +727,7 @@ mod tests {
         let mut bar = StatusBar::new(
             &Theme::default(),
             StatusLineSegment::DEFAULT.to_vec(),
-            "Opus 4.7".into(),
+            "Opus 4.8".into(),
             Some(Effort::Xhigh),
             cwd.into(),
             None,
@@ -814,7 +814,7 @@ mod tests {
                 StatusLineSegment::Model,
                 StatusLineSegment::CurrentDir,
             ],
-            "Opus 4.7".into(),
+            "Opus 4.8".into(),
             Some(Effort::Xhigh),
             "~/projects/demo".into(),
             None,
@@ -822,7 +822,7 @@ mod tests {
         );
         let output = render_top_row(&mut bar, 120);
         let state_at = output.find("Ready").unwrap();
-        let model_at = output.find("Opus 4.7").unwrap();
+        let model_at = output.find("Opus 4.8").unwrap();
         let cwd_at = output.find("~/projects/demo").unwrap();
         assert!(state_at < model_at, "run state should lead: {output:?}");
         assert!(model_at < cwd_at, "cwd should follow model: {output:?}");
@@ -842,14 +842,14 @@ mod tests {
                 StatusLineSegment::ModelWithEffort,
                 StatusLineSegment::RunState,
             ],
-            "Opus 4.7".into(),
+            "Opus 4.8".into(),
             Some(Effort::Xhigh),
             "~/projects/demo".into(),
             None,
             Some("feat/status-line".to_owned()),
         );
         let output = render_top_row(&mut bar, 120);
-        assert!(output.contains("~/projects/demo │ feat/status-line │ Opus 4.7 (xhigh) │ Ready"));
+        assert!(output.contains("~/projects/demo │ feat/status-line │ Opus 4.8 (xhigh) │ Ready"));
     }
 
     #[test]
diff --git a/crates/oxide-code/src/tui/components/welcome.rs b/crates/oxide-code/src/tui/components/welcome.rs
index e1f2d91c..b0bf4adf 100644
--- a/crates/oxide-code/src/tui/components/welcome.rs
+++ b/crates/oxide-code/src/tui/components/welcome.rs
@@ -353,7 +353,7 @@ mod tests {
                 auth_label: "OAuth",
                 base_url: "https://api.test.invalid".to_owned(),
                 extra_ca_certs: None,
-                model_id: "claude-opus-4-7".to_owned(),
+                model_id: "claude-opus-4-8".to_owned(),
                 effort: Some(Effort::Xhigh),
                 max_tokens: 64_000,
                 max_tool_rounds: None,
@@ -392,7 +392,7 @@ mod tests {
         let info = fixture();
         let snap = snap_for(&info);
         assert_eq!(snap.version, "0.1.0");
-        assert_eq!(snap.model_label, "Claude Opus 4.7");
+        assert_eq!(snap.model_label, "Claude Opus 4.8");
         assert_eq!(snap.effort_label, "xhigh");
         assert_eq!(snap.auth_label, "OAuth");
         assert_eq!(snap.cwd, "~/github/oxide-code");
diff --git a/docs/design/tui/status-line.md b/docs/design/tui/status-line.md
index 819bb6d7..7023703c 100644
--- a/docs/design/tui/status-line.md
+++ b/docs/design/tui/status-line.md
@@ -47,7 +47,7 @@ Segment render rules:
 - Use active theme styles; color customization belongs in theme overrides.
 - Join only rendered segments, so omitted segments do not leave extra separators.
 - When the row is too narrow, omit lower-utility segments before truncating the last remaining segment. Run state and model have the highest utility.
-- The `model` and `model-with-effort` segments use a width-tightened label: the `Claude` family prefix is dropped and `(1M context)` becomes `[1M]` (e.g., `Opus 4.7 [1M] (xhigh)`). Other surfaces (welcome screen, prompt environment, error blocks) keep the full `Claude Opus 4.7 (1M context)` label.
+- The `model` and `model-with-effort` segments use a width-tightened label: the `Claude` family prefix is dropped and `(1M context)` becomes `[1M]` (e.g., `Opus 4.8 [1M] (xhigh)`). Other surfaces (welcome screen, prompt environment, error blocks) keep the full `Claude Opus 4.8 (1M context)` label.
 
 ## Usage Data
 
diff --git a/docs/design/tui/welcome.md b/docs/design/tui/welcome.md
index 2aa0fc73..209cd2b5 100644
--- a/docs/design/tui/welcome.md
+++ b/docs/design/tui/welcome.md
@@ -21,12 +21,12 @@ The minimal `Welcome to ox / Ask anything to begin.` banner gets users to the pr
 
 ## Layout
 
-At 80 cols with the test fixture (`claude-opus-4-7` plain, `xhigh` effort, OAuth), the render is:
+At 80 cols with the test fixture (`claude-opus-4-8` plain, `xhigh` effort, OAuth), the render is:
 
 ```text
                            ━━━━ oxide-code v0.1.0 ━━━━
 
-                  Claude Opus 4.7 · xhigh effort · OAuth
+                  Claude Opus 4.8 · xhigh effort · OAuth
                   ~/github/oxide-code
 
                   Try one of:
diff --git a/docs/guide/configuration.md b/docs/guide/configuration.md
index 47e96bc0..c255d563 100644
--- a/docs/guide/configuration.md
+++ b/docs/guide/configuration.md
@@ -45,7 +45,7 @@ status_line = [
 | `api_key`          | string  | -                           | Anthropic API key (user config only)                      |
 | `base_url`         | string  | `https://api.anthropic.com` | API base URL (user config only)                           |
 | `extra_ca_certs`   | string  | -                           | PEM bundle appended to the trust store (user config only) |
-| `model`            | string  | `claude-opus-4-7[1m]`       | Model to use                                              |
+| `model`            | string  | `claude-opus-4-8[1m]`       | Model to use                                              |
 | `effort`           | string  | per-model (see below)       | Intelligence-vs-latency tier                              |
 | `max_tokens`       | integer | effort-derived (see below)  | Max tokens per response                                   |
 | `max_tool_rounds`  | integer | unset (unbounded)           | Per-turn safety cap on tool rounds                        |
@@ -59,15 +59,15 @@ Per-model defaults when `effort` is unset:
 
 | Model           | Default |
 | --------------- | ------- |
-| Opus 4.7        | `xhigh` |
+| Opus 4.8 / 4.7  | `xhigh` |
 | Opus 4.6        | `high`  |
 | Sonnet 4.6      | `high`  |
 | Everything else | (unset) |
 
-Tier guide (from the [Opus 4.7 migration guide](https://platform.claude.com/docs/en/about-claude/models/migration-guide)):
+Tier guide (from the [Opus 4.8 migration guide](https://platform.claude.com/docs/en/about-claude/models/migration-guide)):
 
 - `max`: Deepest reasoning, Opus-only, with diminishing returns on some tasks.
-- `xhigh`: Recommended default for coding and agentic work on Opus 4.7.
+- `xhigh`: Recommended default for coding and agentic work on Opus 4.7 / 4.8.
 - `high`: Balanced, minimum recommended for intelligence-sensitive tasks.
 - `medium`: Cost-sensitive workloads.
 - `low`: Scoped, latency-sensitive tasks.
@@ -120,7 +120,7 @@ Append `[1m]` to `model` to opt into the 1M-token context window on models that
 
 ```toml
 [client]
-model = "claude-opus-4-7[1m]"
+model = "claude-opus-4-8[1m]"
 ```
 
 1M access depends on your subscription or gateway, so you have to opt in explicitly. The tag is silently ignored on models without 1M support (e.g. Haiku).
@@ -133,7 +133,7 @@ model = "claude-opus-4-7[1m]"
 | `show_welcome`  | boolean | `true`    | Paint the welcome splash on an empty chat |
 | `status_line`   | array   | see below | Ordered status-line segments              |
 
-On Opus 4.7, `show_thinking = true` opts the request into `thinking.display = "summarized"` so the API streams reasoning text. Otherwise the 4.7 default of `"omitted"` applies and the UI sees nothing until the final answer arrives.
+On Opus 4.7 / 4.8, `show_thinking = true` opts the request into `thinking.display = "summarized"` so the API streams reasoning text. Otherwise the `"omitted"` default applies and the UI sees nothing until the final answer arrives.
 
 `show_welcome = false` blanks the chat region until you submit a prompt, which is useful when piping or screen-recording.
 
@@ -144,7 +144,7 @@ On Opus 4.7, `show_thinking = true` opts the request into `thinking.display = "s
 | `current-dir`       | Tildified working directory                    | At startup      |
 | `git-branch`        | Current branch (omitted on detached HEAD)      | Every 5 s       |
 | `pull-request`      | Open PR as clickable `#86` when OSC 8 works    | Every 60 s      |
-| `model`             | Compact model label (e.g., `Opus 4.7`)         | On `/model`     |
+| `model`             | Compact model label (e.g., `Opus 4.8`)         | On `/model`     |
 | `model-with-effort` | Model label plus the effort tier in parens     | On `/model`     |
 | `context-used`      | `Ctx: 50% (100k/200k)` after the first turn    | Per turn        |
 | `session-cost`      | `Sess: $0.4321` running USD estimate           | Per turn        |
@@ -196,7 +196,7 @@ Environment variables override all config file values.
 | -------------------------------------- | ------------------------------------------ | --------------------------- | ---------------------------- |
 | `ANTHROPIC_API_KEY`                    | `client.api_key`                           | -                           | Anthropic API key            |
 | `ANTHROPIC_BASE_URL`                   | `client.base_url`                          | `https://api.anthropic.com` | API base URL                 |
-| `ANTHROPIC_MODEL`                      | `client.model`                             | `claude-opus-4-7[1m]`       | Model to use                 |
+| `ANTHROPIC_MODEL`                      | `client.model`                             | `claude-opus-4-8[1m]`       | Model to use                 |
 | `ANTHROPIC_EFFORT`                     | `client.effort`                            | per-model                   | Intelligence-vs-latency tier |
 | `ANTHROPIC_MAX_TOKENS`                 | `client.max_tokens`                        | effort-derived              | Max tokens per response      |
 | `OX_EXTRA_CA_CERTS`                    | `client.extra_ca_certs`                    | -                           | Path to a PEM trust bundle   |
diff --git a/docs/research/api/anthropic.md b/docs/research/api/anthropic.md
index ee50c96e..7ca29f2d 100644
--- a/docs/research/api/anthropic.md
+++ b/docs/research/api/anthropic.md
@@ -85,7 +85,7 @@ Key rules:
 - **Haiku + `interleaved-thinking`**: Third-party gateways reject it, while first-party accepts.
 - **Haiku one-shots** (title generation, compaction classifier): Strip agentic markers entirely. `claude-code-20250219` is re-added only when the call is agentic.
 - **`prompt-caching-scope` ships unconditionally.** The header alone is a server-side no-op without a matching `cache_control.scope` field, but 3P gateways fingerprint its absence. oxide-code therefore emits the beta on every agentic request and gates only the body-side `cache_control.scope: "global"` on `is_first_party_base_url()` (see [Prompt Caching Scope](#prompt-caching-scope) for why).
-- **`context-1m` is user opt-in via `[1m]`.** Appending `[1m]` to the model string (e.g., `claude-opus-4-7[1m]`) adds the 1M beta and strips the tag before the request hits the wire. Family-based auto-enable would 400 on subscriptions or gateways that do not carry 1M access. Convention matches Claude Code.
+- **`context-1m` is user opt-in via `[1m]`.** Appending `[1m]` to the model string (e.g., `claude-opus-4-8[1m]`) adds the 1M beta and strips the tag before the request hits the wire. Family-based auto-enable would 400 on subscriptions or gateways that do not carry 1M access. Convention matches Claude Code.
 - **`effort` is Opus 4.6+ and Sonnet 4.6+ only.** Opus 4.5 and older, Sonnet 4.5 and older, and all Haiku variants reject it per upstream's `modelSupportsEffort`. Per-model support is encoded in `Capabilities::supported_efforts`; `accepts_effort`, `clamp_effort`, and `default_effort` keep user picks and defaults inside that set.
 - **`effort` and `context-management` betas need a body field.** Sending the header alone is a silent no-op: the request runs at the server default. See [Agentic Request Body Fields](#agentic-request-body-fields) for the matching `output_config.effort` and `context_management.edits` shapes. oxide-code pairs each capability with both its beta and its body field so the two stay in sync.
 - **`structured-outputs` is per-version and caller-opt-in.** The upstream allowlist is Opus 4.1 / 4.5 / 4.6+, Sonnet 4.5 / 4.6+, Haiku 4.5. The beta ships only when a caller supplies an `output_config.format` (today: the AI-title generator). The body field and header are paired on the same capability flag. A schema passed to an unsupported model silently falls back to free-form text, mirroring the `[1m]` × `context_1m` silent-strip pattern.
@@ -237,8 +237,8 @@ GA as of Opus 4.6. Controls the intelligence-vs-latency tier of agentic turns vi
 ```
 
 - **The `effort-2025-11-24` beta header is necessary but not sufficient.** oxide-code used to send the header without the body field. The header became a no-op and the model ran at an undefined default.
-- **Per-model ceiling.** `max` is Opus-only, and Sonnet 4.6 400s on it. `xhigh` is Opus 4.7-only. `Capabilities::supported_efforts` encodes the allowed set, and `Capabilities::clamp_effort` clamps a user pick down to the highest supported level at or below it.
-- **Per-model default.** Claude Code 2.1.119 sends `xhigh` on Opus 4.7, `high` on Opus 4.6 and Sonnet 4.6, omits the field entirely on earlier models. oxide-code mirrors this via `Capabilities::default_effort`.
+- **Per-model ceiling.** `max` is Opus-only, and Sonnet 4.6 400s on it. `xhigh` is Opus 4.7 / 4.8-only. `Capabilities::supported_efforts` encodes the allowed set, and `Capabilities::clamp_effort` clamps a user pick down to the highest supported level at or below it.
+- **Per-model default.** `Capabilities::default_effort` picks the highest non-`max` tier: `xhigh` on Opus 4.7 / 4.8, `high` on Opus 4.6 and Sonnet 4.6, omitted on models without effort. This tracks Claude Code for 4.6 / 4.7; oxide-code keeps `xhigh` as the 4.8 default even though upstream lowered the stock 4.8 default to `high`.
 - **`max_tokens` should scale with effort.** Claude Code uses 64 K on Opus 4.7 at `xhigh`, 32 K on Sonnet 4.6 at `high`. oxide-code's `default_max_tokens(effort)` matches the upper tiers and uses 16 K otherwise when the user hasn't set `ANTHROPIC_MAX_TOKENS` explicitly.
 
 ### `context_management.edits`
diff --git a/docs/research/tui/status-line.md b/docs/research/tui/status-line.md
index 066f6358..4ea2ba05 100644
--- a/docs/research/tui/status-line.md
+++ b/docs/research/tui/status-line.md
@@ -39,12 +39,12 @@ opencode keeps usage in the prompt footer rather than a fully configurable statu
 
 Anthropic's pricing page lists first-party Claude API prices in USD per million tokens and separate prompt-cache rates for 5-minute writes, 1-hour writes, cache reads, and output tokens. Checked on 2026-05-14, the relevant rows for oxide-code's model table are:
 
-| Family               | Input | 5m cache write | 1h cache write | Cache read | Output |
-| -------------------- | ----- | -------------- | -------------- | ---------- | ------ |
-| Opus 4.7 / 4.6 / 4.5 | $5    | $6.25          | $10            | $0.50      | $25    |
-| Opus 4.1             | $15   | $18.75         | $30            | $1.50      | $75    |
-| Sonnet 4.x           | $3    | $3.75          | $6             | $0.30      | $15    |
-| Haiku 4.5            | $1    | $1.25          | $2             | $0.10      | $5     |
+| Family                     | Input | 5m cache write | 1h cache write | Cache read | Output |
+| -------------------------- | ----- | -------------- | -------------- | ---------- | ------ |
+| Opus 4.8 / 4.7 / 4.6 / 4.5 | $5    | $6.25          | $10            | $0.50      | $25    |
+| Opus 4.1                   | $15   | $18.75         | $30            | $1.50      | $75    |
+| Sonnet 4.x                 | $3    | $3.75          | $6             | $0.30      | $15    |
+| Haiku 4.5                  | $1    | $1.25          | $2             | $0.10      | $5     |
 
 Cost display should stay best-effort because account discounts, marketplace billing, data residency, fast mode, and server-side tool pricing can change the final bill.
 

From 485f5b48f9e88447e4f25d097d347850113a2131 Mon Sep 17 00:00:00 2001
From: Hakula Chen <i@hakula.xyz>
Date: Fri, 29 May 2026 16:56:41 +0800
Subject: [PATCH 2/5] refactor(model): drop retired Opus 4.1

Opus 4.1 is no longer a supported model. Remove its MODELS row, the
OPUS_4_1_RATES pricing constant, and the tests that pinned its older
$15/$75 pricing and display labels. `lookup` now treats `claude-opus-4-1`
as a retired family that resolves to nothing, asserted alongside the other
absent ids.

Collapse the Opus 4.1 references in the API and status-line research docs
into the 4.5 tier, which carries the same beta profile and is the lowest
Opus the table still models.
---
 crates/oxide-code/src/model.rs         | 33 ++------------------------
 crates/oxide-code/src/model/pricing.rs |  8 -------
 crates/oxide-code/src/slash/model.rs   |  1 -
 docs/research/api/anthropic.md         | 22 ++++++++---------
 docs/research/tui/status-line.md       |  1 -
 5 files changed, 13 insertions(+), 52 deletions(-)

diff --git a/crates/oxide-code/src/model.rs b/crates/oxide-code/src/model.rs
index 7abd6a9c..d1d9aa40 100644
--- a/crates/oxide-code/src/model.rs
+++ b/crates/oxide-code/src/model.rs
@@ -7,7 +7,7 @@ use std::borrow::Cow;
 use crate::config::Effort;
 
 pub(crate) use pricing::TokenCostRates;
-use pricing::{HAIKU_RATES, OPUS_4_1_RATES, OPUS_4_5_PLUS_RATES, SONNET_RATES};
+use pricing::{HAIKU_RATES, OPUS_4_5_PLUS_RATES, SONNET_RATES};
 
 // ── ModelInfo ──
 
@@ -155,19 +155,6 @@ pub(crate) const MODELS: &[ModelInfo] = &[
         },
         cost_rates: Some(HAIKU_RATES),
     },
-    ModelInfo {
-        id_substr: "claude-opus-4-1",
-        display_name: "Claude Opus 4.1",
-        cutoff: Some("January 2025"),
-        capabilities: Capabilities {
-            interleaved_thinking: true,
-            context_management: true,
-            context_1m: false,
-            supported_efforts: &[],
-            structured_outputs: true,
-        },
-        cost_rates: Some(OPUS_4_1_RATES),
-    },
 ];
 
 impl Capabilities {
@@ -343,7 +330,6 @@ mod tests {
             "claude-opus-4-5",
             "claude-sonnet-4-5",
             "claude-haiku-4-5",
-            "claude-opus-4-1",
         ] {
             assert!(
                 !lookup(other)
@@ -564,6 +550,7 @@ mod tests {
             "claude-sonnet-4",
             "claude-haiku-4",
             "claude-opus-4-20250514",
+            "claude-opus-4-1",
             "gpt-4",
         ] {
             assert!(lookup(unknown).is_none(), "{unknown} must not resolve");
@@ -613,20 +600,6 @@ mod tests {
         assert!((one_hour - 40.5).abs() < 1e-9);
     }
 
-    #[test]
-    fn token_cost_rates_for_opus_4_1_uses_older_pricing() {
-        let rates = token_cost_rates_for("claude-opus-4-1").unwrap();
-        let cost = rates.estimate_usd(
-            1_000_000,
-            1_000_000,
-            1_000_000,
-            1_000_000,
-            PromptCacheTtl::FiveMin,
-        );
-
-        assert!((cost - 110.25).abs() < 1e-9);
-    }
-
     #[test]
     fn token_cost_rates_for_unknown_model_is_absent() {
         assert!(token_cost_rates_for("claude-future-9").is_none());
@@ -644,7 +617,6 @@ mod tests {
             ("claude-opus-4-5", "Claude Opus 4.5"),
             ("claude-sonnet-4-5", "Claude Sonnet 4.5"),
             ("claude-haiku-4-5", "Claude Haiku 4.5"),
-            ("claude-opus-4-1", "Claude Opus 4.1"),
         ] {
             assert_eq!(display_name(id), expected, "{id}");
         }
@@ -678,7 +650,6 @@ mod tests {
             ("claude-opus-4-7", "Opus 4.7"),
             ("claude-sonnet-4-6", "Sonnet 4.6"),
             ("claude-haiku-4-5", "Haiku 4.5"),
-            ("claude-opus-4-1", "Opus 4.1"),
         ] {
             assert_eq!(short_display_name(id), expected, "{id}");
         }
diff --git a/crates/oxide-code/src/model/pricing.rs b/crates/oxide-code/src/model/pricing.rs
index 50ae6252..08cde7ec 100644
--- a/crates/oxide-code/src/model/pricing.rs
+++ b/crates/oxide-code/src/model/pricing.rs
@@ -24,14 +24,6 @@ pub(super) const OPUS_4_5_PLUS_RATES: TokenCostRates = TokenCostRates {
     output: 25.0,
 };
 
-pub(super) const OPUS_4_1_RATES: TokenCostRates = TokenCostRates {
-    input: 15.0,
-    cache_write_5m: 18.75,
-    cache_write_1h: 30.0,
-    cache_read: 1.50,
-    output: 75.0,
-};
-
 pub(super) const SONNET_RATES: TokenCostRates = TokenCostRates {
     input: 3.0,
     cache_write_5m: 3.75,
diff --git a/crates/oxide-code/src/slash/model.rs b/crates/oxide-code/src/slash/model.rs
index c642724a..fc976756 100644
--- a/crates/oxide-code/src/slash/model.rs
+++ b/crates/oxide-code/src/slash/model.rs
@@ -470,7 +470,6 @@ mod tests {
         for dated in [
             "claude-opus-4-7-20260101",
             "claude-opus-4-6-20250805",
-            "claude-opus-4-1-20250805",
             "claude-sonnet-4-5-20250929",
         ] {
             assert_eq!(
diff --git a/docs/research/api/anthropic.md b/docs/research/api/anthropic.md
index 7ca29f2d..a9b192c4 100644
--- a/docs/research/api/anthropic.md
+++ b/docs/research/api/anthropic.md
@@ -68,16 +68,16 @@ Rows are grouped by role: identity / auth → universal agentic → model-tier-g
 
 Cell legend: `✓` always on, `-` not supported (or stripped), `[1m]` opt-in via the model suffix, `*` caller opt-in (body field + beta ship together, see rules below).
 
-| Beta                              | Opus 4 (base) | Opus 4.1 / 4.5 | Opus 4.6+ | Sonnet 4 (base) | Sonnet 4.5 | Sonnet 4.6+ | Haiku 4 (base) | Haiku 4.5 (agentic) | Haiku 4.5 (one-shot) |
-| --------------------------------- | ------------- | -------------- | --------- | --------------- | ---------- | ----------- | -------------- | ------------------- | -------------------- |
-| `claude-code-20250219`            | ✓             | ✓              | ✓         | ✓               | ✓          | ✓           | ✓              | ✓                   | -                    |
-| `oauth-2025-04-20` (OAuth only)   | ✓             | ✓              | ✓         | ✓               | ✓          | ✓           | ✓              | ✓                   | ✓                    |
-| `context-management-2025-06-27`   | ✓             | ✓              | ✓         | ✓               | ✓          | ✓           | ✓              | ✓                   | -                    |
-| `prompt-caching-scope-2026-01-05` | ✓             | ✓              | ✓         | ✓               | ✓          | ✓           | ✓              | ✓                   | -                    |
-| `interleaved-thinking-2025-05-14` | ✓             | ✓              | ✓         | ✓               | ✓          | ✓           | -              | -                   | -                    |
-| `context-1m-2025-08-07`           | -             | -              | `[1m]`    | `[1m]`          | `[1m]`     | `[1m]`      | -              | -                   | -                    |
-| `effort-2025-11-24`               | -             | -              | ✓         | -               | -          | ✓           | -              | -                   | -                    |
-| `structured-outputs-2025-12-15`   | -             | `*`            | `*`       | -               | `*`        | `*`         | -              | `*`                 | `*`                  |
+| Beta                              | Opus 4 (base) | Opus 4.5 | Opus 4.6+ | Sonnet 4 (base) | Sonnet 4.5 | Sonnet 4.6+ | Haiku 4 (base) | Haiku 4.5 (agentic) | Haiku 4.5 (one-shot) |
+| --------------------------------- | ------------- | -------- | --------- | --------------- | ---------- | ----------- | -------------- | ------------------- | -------------------- |
+| `claude-code-20250219`            | ✓             | ✓        | ✓         | ✓               | ✓          | ✓           | ✓              | ✓                   | -                    |
+| `oauth-2025-04-20` (OAuth only)   | ✓             | ✓        | ✓         | ✓               | ✓          | ✓           | ✓              | ✓                   | ✓                    |
+| `context-management-2025-06-27`   | ✓             | ✓        | ✓         | ✓               | ✓          | ✓           | ✓              | ✓                   | -                    |
+| `prompt-caching-scope-2026-01-05` | ✓             | ✓        | ✓         | ✓               | ✓          | ✓           | ✓              | ✓                   | -                    |
+| `interleaved-thinking-2025-05-14` | ✓             | ✓        | ✓         | ✓               | ✓          | ✓           | -              | -                   | -                    |
+| `context-1m-2025-08-07`           | -             | -        | `[1m]`    | `[1m]`          | `[1m]`     | `[1m]`      | -              | -                   | -                    |
+| `effort-2025-11-24`               | -             | -        | ✓         | -               | -          | ✓           | -              | -                   | -                    |
+| `structured-outputs-2025-12-15`   | -             | `*`      | `*`       | -               | `*`        | `*`         | -              | `*`                 | `*`                  |
 
 Key rules:
 
@@ -88,7 +88,7 @@ Key rules:
 - **`context-1m` is user opt-in via `[1m]`.** Appending `[1m]` to the model string (e.g., `claude-opus-4-8[1m]`) adds the 1M beta and strips the tag before the request hits the wire. Family-based auto-enable would 400 on subscriptions or gateways that do not carry 1M access. Convention matches Claude Code.
 - **`effort` is Opus 4.6+ and Sonnet 4.6+ only.** Opus 4.5 and older, Sonnet 4.5 and older, and all Haiku variants reject it per upstream's `modelSupportsEffort`. Per-model support is encoded in `Capabilities::supported_efforts`; `accepts_effort`, `clamp_effort`, and `default_effort` keep user picks and defaults inside that set.
 - **`effort` and `context-management` betas need a body field.** Sending the header alone is a silent no-op: the request runs at the server default. See [Agentic Request Body Fields](#agentic-request-body-fields) for the matching `output_config.effort` and `context_management.edits` shapes. oxide-code pairs each capability with both its beta and its body field so the two stay in sync.
-- **`structured-outputs` is per-version and caller-opt-in.** The upstream allowlist is Opus 4.1 / 4.5 / 4.6+, Sonnet 4.5 / 4.6+, Haiku 4.5. The beta ships only when a caller supplies an `output_config.format` (today: the AI-title generator). The body field and header are paired on the same capability flag. A schema passed to an unsupported model silently falls back to free-form text, mirroring the `[1m]` × `context_1m` silent-strip pattern.
+- **`structured-outputs` is per-version and caller-opt-in.** The upstream allowlist is Opus 4.5 / 4.6+, Sonnet 4.5 / 4.6+, Haiku 4.5. The beta ships only when a caller supplies an `output_config.format` (today: the AI-title generator). The body field and header are paired on the same capability flag. A schema passed to an unsupported model silently falls back to free-form text, mirroring the `[1m]` × `context_1m` silent-strip pattern.
 - **Unknown model aliases** fall through substring matching on the family stem. `claude-opus-5-x` would miss every row and ship with only the identity / caching betas. Bump the `MODELS` table when a new family lands.
 
 oxide-code gates each beta header on the target model in `client::anthropic::compute_betas`, which consults the ground-truth `Capabilities` flags in `crate::model::MODELS`. New models ship by adding a row to that table, with no beta-logic changes needed.
diff --git a/docs/research/tui/status-line.md b/docs/research/tui/status-line.md
index 4ea2ba05..568b91d8 100644
--- a/docs/research/tui/status-line.md
+++ b/docs/research/tui/status-line.md
@@ -42,7 +42,6 @@ Anthropic's pricing page lists first-party Claude API prices in USD per million
 | Family                     | Input | 5m cache write | 1h cache write | Cache read | Output |
 | -------------------------- | ----- | -------------- | -------------- | ---------- | ------ |
 | Opus 4.8 / 4.7 / 4.6 / 4.5 | $5    | $6.25          | $10            | $0.50      | $25    |
-| Opus 4.1                   | $15   | $18.75         | $30            | $1.50      | $75    |
 | Sonnet 4.x                 | $3    | $3.75          | $6             | $0.30      | $15    |
 | Haiku 4.5                  | $1    | $1.25          | $2             | $0.10      | $5     |
 

From 2726b38b79df6a16084711e4f0f87679cfda5857 Mon Sep 17 00:00:00 2001
From: Hakula Chen <i@hakula.xyz>
Date: Fri, 29 May 2026 17:03:36 +0800
Subject: [PATCH 3/5] docs(research): drop unmodeled base-4 columns from the
 beta matrix

The beta matrix mirrors the families oxide-code's MODELS table actually
sends. Base Opus 4, Sonnet 4, and Haiku 4 resolve to nothing in `lookup`
(asserted by lookup_unknown_or_retired), so their columns documented
profiles the client never emits. The 4.5 vs 4.6+ columns still carry the
tier-gating contrast (effort, context-1m), and the unknown-alias fallback
is already covered in prose.
---
 docs/research/api/anthropic.md | 20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)

diff --git a/docs/research/api/anthropic.md b/docs/research/api/anthropic.md
index a9b192c4..c0cfa5c5 100644
--- a/docs/research/api/anthropic.md
+++ b/docs/research/api/anthropic.md
@@ -68,16 +68,16 @@ Rows are grouped by role: identity / auth → universal agentic → model-tier-g
 
 Cell legend: `✓` always on, `-` not supported (or stripped), `[1m]` opt-in via the model suffix, `*` caller opt-in (body field + beta ship together, see rules below).
 
-| Beta                              | Opus 4 (base) | Opus 4.5 | Opus 4.6+ | Sonnet 4 (base) | Sonnet 4.5 | Sonnet 4.6+ | Haiku 4 (base) | Haiku 4.5 (agentic) | Haiku 4.5 (one-shot) |
-| --------------------------------- | ------------- | -------- | --------- | --------------- | ---------- | ----------- | -------------- | ------------------- | -------------------- |
-| `claude-code-20250219`            | ✓             | ✓        | ✓         | ✓               | ✓          | ✓           | ✓              | ✓                   | -                    |
-| `oauth-2025-04-20` (OAuth only)   | ✓             | ✓        | ✓         | ✓               | ✓          | ✓           | ✓              | ✓                   | ✓                    |
-| `context-management-2025-06-27`   | ✓             | ✓        | ✓         | ✓               | ✓          | ✓           | ✓              | ✓                   | -                    |
-| `prompt-caching-scope-2026-01-05` | ✓             | ✓        | ✓         | ✓               | ✓          | ✓           | ✓              | ✓                   | -                    |
-| `interleaved-thinking-2025-05-14` | ✓             | ✓        | ✓         | ✓               | ✓          | ✓           | -              | -                   | -                    |
-| `context-1m-2025-08-07`           | -             | -        | `[1m]`    | `[1m]`          | `[1m]`     | `[1m]`      | -              | -                   | -                    |
-| `effort-2025-11-24`               | -             | -        | ✓         | -               | -          | ✓           | -              | -                   | -                    |
-| `structured-outputs-2025-12-15`   | -             | `*`      | `*`       | -               | `*`        | `*`         | -              | `*`                 | `*`                  |
+| Beta                              | Opus 4.5 | Opus 4.6+ | Sonnet 4.5 | Sonnet 4.6+ | Haiku 4.5 (agentic) | Haiku 4.5 (one-shot) |
+| --------------------------------- | -------- | --------- | ---------- | ----------- | ------------------- | -------------------- |
+| `claude-code-20250219`            | ✓        | ✓         | ✓          | ✓           | ✓                   | -                    |
+| `oauth-2025-04-20` (OAuth only)   | ✓        | ✓         | ✓          | ✓           | ✓                   | ✓                    |
+| `context-management-2025-06-27`   | ✓        | ✓         | ✓          | ✓           | ✓                   | -                    |
+| `prompt-caching-scope-2026-01-05` | ✓        | ✓         | ✓          | ✓           | ✓                   | -                    |
+| `interleaved-thinking-2025-05-14` | ✓        | ✓         | ✓          | ✓           | -                   | -                    |
+| `context-1m-2025-08-07`           | -        | `[1m]`    | `[1m]`     | `[1m]`      | -                   | -                    |
+| `effort-2025-11-24`               | -        | ✓         | -          | ✓           | -                   | -                    |
+| `structured-outputs-2025-12-15`   | `*`      | `*`       | `*`        | `*`         | `*`                 | `*`                  |
 
 Key rules:
 

From 29265e43e2c202e331dde236c841913c1312215a Mon Sep 17 00:00:00 2001
From: Hakula Chen <i@hakula.xyz>
Date: Fri, 29 May 2026 17:40:22 +0800
Subject: [PATCH 4/5] feat(model): default effort per model, mirroring Claude
 Code

The highest-non-Max heuristic resolved both Opus 4.7 and 4.8 to xhigh, but
upstream defaults every effort-capable model to `high` and Claude Code bumps
only 4.7 to `xhigh` (4.8 rides the `high` default). Replace the heuristic with
an explicit per-model `default_effort` on Capabilities: xhigh on 4.7, high on
4.8 / 4.6 / Sonnet 4.6, none on models without effort. The default model
(Opus 4.8) now resolves to `high`, so derived max_tokens is 32K.
---
 crates/oxide-code/src/config.rs | 14 ++++----
 crates/oxide-code/src/model.rs  | 57 ++++++++++++++++-----------------
 2 files changed, 33 insertions(+), 38 deletions(-)

diff --git a/crates/oxide-code/src/config.rs b/crates/oxide-code/src/config.rs
index df31713c..210b36cf 100644
--- a/crates/oxide-code/src/config.rs
+++ b/crates/oxide-code/src/config.rs
@@ -879,15 +879,15 @@ mod tests {
 
     #[tokio::test]
     async fn load_defaults_apply_when_no_config_and_no_env() {
-        // Opus 4.8 supports `xhigh`; `effort` / `max_tokens` derive from that ceiling.
+        // Opus 4.8 defaults to `high` (mirroring Claude Code); `max_tokens` derives from that.
         let dir = tempfile::tempdir().unwrap();
         let config = temp_env::async_with_vars(env_vars(vec![xdg(&dir)]), Config::load())
             .await
             .unwrap();
         assert_eq!(config.model, DEFAULT_MODEL);
         assert_eq!(config.base_url, DEFAULT_BASE_URL);
-        assert_eq!(config.max_tokens, 64_000);
-        assert_eq!(config.effort, Some(Effort::Xhigh));
+        assert_eq!(config.max_tokens, 32_000);
+        assert_eq!(config.effort, Some(Effort::High));
         assert_eq!(config.prompt_cache_ttl, PromptCacheTtl::OneHour);
         assert!(config.compaction.auto.enabled);
         assert_eq!(
@@ -1505,13 +1505,11 @@ mod tests {
     // ── Config::load / effort resolution ──
 
     #[tokio::test]
-    async fn load_effort_default_follows_model_ceiling() {
+    async fn load_effort_defaults_to_per_model_value() {
+        // 4.7 ships xhigh, 4.8 rides the high default; no-effort models stay unset.
         for (model, expected) in [
-            ("claude-opus-4-8", Some(Effort::Xhigh)),
             ("claude-opus-4-7", Some(Effort::Xhigh)),
-            ("claude-opus-4-6", Some(Effort::High)),
-            ("claude-sonnet-4-6", Some(Effort::High)),
-            ("claude-sonnet-4-5", None),
+            ("claude-opus-4-8", Some(Effort::High)),
             ("claude-haiku-4-5", None),
         ] {
             let dir = tempfile::tempdir().unwrap();
diff --git a/crates/oxide-code/src/model.rs b/crates/oxide-code/src/model.rs
index d1d9aa40..311898b1 100644
--- a/crates/oxide-code/src/model.rs
+++ b/crates/oxide-code/src/model.rs
@@ -43,6 +43,10 @@ pub(crate) struct Capabilities {
     pub(crate) context_1m: bool,
     /// `output_config.effort` levels accepted upstream. Empty when the model rejects `effort`.
     pub(crate) supported_efforts: &'static [Effort],
+    /// Effort sent when the user picks none. Mirrors Claude Code's per-model stock default, which
+    /// differs from a model's ceiling (Opus 4.7 ships `xhigh`, Opus 4.8 rides the `high` default).
+    /// `None` when the model rejects `effort`.
+    pub(crate) default_effort: Option<Effort>,
     /// `structured-outputs-2025-12-15` beta.
     pub(crate) structured_outputs: bool,
 }
@@ -66,6 +70,7 @@ pub(crate) const MODELS: &[ModelInfo] = &[
                 Effort::Xhigh,
                 Effort::Max,
             ],
+            default_effort: Some(Effort::High),
             structured_outputs: true,
         },
         cost_rates: Some(OPUS_4_5_PLUS_RATES),
@@ -85,6 +90,7 @@ pub(crate) const MODELS: &[ModelInfo] = &[
                 Effort::Xhigh,
                 Effort::Max,
             ],
+            default_effort: Some(Effort::Xhigh),
             structured_outputs: true,
         },
         cost_rates: Some(OPUS_4_5_PLUS_RATES),
@@ -98,6 +104,7 @@ pub(crate) const MODELS: &[ModelInfo] = &[
             context_management: true,
             context_1m: true,
             supported_efforts: &[Effort::Low, Effort::Medium, Effort::High, Effort::Max],
+            default_effort: Some(Effort::High),
             structured_outputs: true,
         },
         cost_rates: Some(OPUS_4_5_PLUS_RATES),
@@ -111,6 +118,7 @@ pub(crate) const MODELS: &[ModelInfo] = &[
             context_management: true,
             context_1m: true,
             supported_efforts: &[Effort::Low, Effort::Medium, Effort::High],
+            default_effort: Some(Effort::High),
             structured_outputs: true,
         },
         cost_rates: Some(SONNET_RATES),
@@ -124,6 +132,7 @@ pub(crate) const MODELS: &[ModelInfo] = &[
             context_management: true,
             context_1m: false,
             supported_efforts: &[],
+            default_effort: None,
             structured_outputs: true,
         },
         cost_rates: Some(OPUS_4_5_PLUS_RATES),
@@ -137,6 +146,7 @@ pub(crate) const MODELS: &[ModelInfo] = &[
             context_management: true,
             context_1m: true,
             supported_efforts: &[],
+            default_effort: None,
             structured_outputs: true,
         },
         cost_rates: Some(SONNET_RATES),
@@ -151,6 +161,7 @@ pub(crate) const MODELS: &[ModelInfo] = &[
             context_management: true,
             context_1m: false,
             supported_efforts: &[],
+            default_effort: None,
             structured_outputs: true,
         },
         cost_rates: Some(HAIKU_RATES),
@@ -177,21 +188,11 @@ impl Capabilities {
             .find(|&level| level <= pick)
     }
 
-    /// Default tier when the user hasn't picked one. `Max` is opt-in, so the implicit ceiling is
-    /// the highest non-`Max` supported level.
-    pub(crate) fn default_effort(self) -> Option<Effort> {
-        self.supported_efforts
-            .iter()
-            .copied()
-            .rev()
-            .find(|&level| level != Effort::Max)
-    }
-
-    /// Clamps `pick` when present, otherwise falls back to [`Self::default_effort`].
+    /// Clamps `pick` when present, otherwise falls back to the model's stock `default_effort`.
     pub(crate) fn resolve_effort(self, pick: Option<Effort>) -> Option<Effort> {
         match pick {
             Some(p) => self.clamp_effort(p),
-            None => self.default_effort(),
+            None => self.default_effort,
         }
     }
 }
@@ -463,22 +464,16 @@ mod tests {
     // ── Capabilities::default_effort ──
 
     #[test]
-    fn default_effort_picks_highest_supported_tier_when_user_has_no_pick() {
-        // Opus 4.7 / 4.8: full ladder → xhigh.
-        for id in ["claude-opus-4-7", "claude-opus-4-8"] {
-            let caps = lookup(id).unwrap().capabilities;
-            assert_eq!(caps.default_effort(), Some(Effort::Xhigh), "{id}");
-        }
-
-        // Opus 4.6 / Sonnet 4.6: effort but no xhigh → high.
-        for id in ["claude-opus-4-6", "claude-sonnet-4-6"] {
-            let caps = lookup(id).unwrap().capabilities;
-            assert_eq!(caps.default_effort(), Some(Effort::High), "{id}");
+    fn default_effort_stays_within_supported_efforts() {
+        // Each row's stock default must be a tier the model accepts (or `None` when it rejects
+        // effort), so resolve_effort never emits a level the API would 400 on.
+        for info in MODELS {
+            let caps = info.capabilities;
+            match caps.default_effort {
+                Some(level) => assert!(caps.accepts_effort(level), "{}", info.id_substr),
+                None => assert!(!caps.has_effort(), "{}", info.id_substr),
+            }
         }
-
-        // No effort tier at all → None.
-        let haiku_4_5 = lookup("claude-haiku-4-5").unwrap().capabilities;
-        assert_eq!(haiku_4_5.default_effort(), None);
     }
 
     // ── Capabilities::resolve_effort ──
@@ -503,11 +498,13 @@ mod tests {
     }
 
     #[test]
-    fn resolve_effort_falls_back_to_model_default_when_pick_is_none() {
+    fn resolve_effort_falls_back_to_per_model_default_when_pick_is_none() {
+        // 4.7 and 4.8 share the full ladder but default differently: 4.7 ships xhigh, 4.8 rides
+        // the high default (mirrors Claude Code).
         let opus_4_7 = lookup("claude-opus-4-7").unwrap().capabilities;
         assert_eq!(opus_4_7.resolve_effort(None), Some(Effort::Xhigh));
-        let sonnet_4_6 = lookup("claude-sonnet-4-6").unwrap().capabilities;
-        assert_eq!(sonnet_4_6.resolve_effort(None), Some(Effort::High));
+        let opus_4_8 = lookup("claude-opus-4-8").unwrap().capabilities;
+        assert_eq!(opus_4_8.resolve_effort(None), Some(Effort::High));
     }
 
     #[test]

From ccf1789b4f210ce8b98efc2153c7ffb98c73a1df Mon Sep 17 00:00:00 2001
From: Hakula Chen <i@hakula.xyz>
Date: Fri, 29 May 2026 17:40:22 +0800
Subject: [PATCH 5/5] docs: keep upstream model coverage and fix effort
 references

The API and status-line research docs describe upstream rather than
oxide-code's supported set, so restore Opus 4.1 and the base-4 columns in the
beta matrix and pricing table (oxide-code still drops 4.1 from its own model
table). Collapse contiguous version ranges to the `4.1+` / `4.5+` shorthand.

Correct the effort docs: the per-model default note now states the real
upstream default (`high` everywhere, with Claude Code bumping 4.7 to `xhigh`),
the configuration guide's default table matches, and the tier-guide link
points to the effort doc that carries the table instead of a mislabeled
migration-guide link.
---
 docs/guide/configuration.md      |  8 +++++---
 docs/research/api/anthropic.md   | 24 ++++++++++++------------
 docs/research/tui/status-line.md | 11 ++++++-----
 3 files changed, 23 insertions(+), 20 deletions(-)

diff --git a/docs/guide/configuration.md b/docs/guide/configuration.md
index c255d563..f6e04f54 100644
--- a/docs/guide/configuration.md
+++ b/docs/guide/configuration.md
@@ -59,12 +59,14 @@ Per-model defaults when `effort` is unset:
 
 | Model           | Default |
 | --------------- | ------- |
-| Opus 4.8 / 4.7  | `xhigh` |
-| Opus 4.6        | `high`  |
+| Opus 4.7        | `xhigh` |
+| Opus 4.8 / 4.6  | `high`  |
 | Sonnet 4.6      | `high`  |
 | Everything else | (unset) |
 
-Tier guide (from the [Opus 4.8 migration guide](https://platform.claude.com/docs/en/about-claude/models/migration-guide)):
+oxide-code mirrors Claude Code here: `high` is the upstream API default, and Opus 4.7 ships `xhigh`. `xhigh` stays the recommended pick for coding on both 4.7 and 4.8, so set it explicitly (or via `/effort`) when you want it on 4.8.
+
+Tier guide (from Anthropic's [effort documentation](https://platform.claude.com/docs/en/build-with-claude/effort)):
 
 - `max`: Deepest reasoning, Opus-only, with diminishing returns on some tasks.
 - `xhigh`: Recommended default for coding and agentic work on Opus 4.7 / 4.8.
diff --git a/docs/research/api/anthropic.md b/docs/research/api/anthropic.md
index c0cfa5c5..1c676e78 100644
--- a/docs/research/api/anthropic.md
+++ b/docs/research/api/anthropic.md
@@ -68,16 +68,16 @@ Rows are grouped by role: identity / auth → universal agentic → model-tier-g
 
 Cell legend: `✓` always on, `-` not supported (or stripped), `[1m]` opt-in via the model suffix, `*` caller opt-in (body field + beta ship together, see rules below).
 
-| Beta                              | Opus 4.5 | Opus 4.6+ | Sonnet 4.5 | Sonnet 4.6+ | Haiku 4.5 (agentic) | Haiku 4.5 (one-shot) |
-| --------------------------------- | -------- | --------- | ---------- | ----------- | ------------------- | -------------------- |
-| `claude-code-20250219`            | ✓        | ✓         | ✓          | ✓           | ✓                   | -                    |
-| `oauth-2025-04-20` (OAuth only)   | ✓        | ✓         | ✓          | ✓           | ✓                   | ✓                    |
-| `context-management-2025-06-27`   | ✓        | ✓         | ✓          | ✓           | ✓                   | -                    |
-| `prompt-caching-scope-2026-01-05` | ✓        | ✓         | ✓          | ✓           | ✓                   | -                    |
-| `interleaved-thinking-2025-05-14` | ✓        | ✓         | ✓          | ✓           | -                   | -                    |
-| `context-1m-2025-08-07`           | -        | `[1m]`    | `[1m]`     | `[1m]`      | -                   | -                    |
-| `effort-2025-11-24`               | -        | ✓         | -          | ✓           | -                   | -                    |
-| `structured-outputs-2025-12-15`   | `*`      | `*`       | `*`        | `*`         | `*`                 | `*`                  |
+| Beta                              | Opus 4 (base) | Opus 4.1 / 4.5 | Opus 4.6+ | Sonnet 4 (base) | Sonnet 4.5 | Sonnet 4.6+ | Haiku 4 (base) | Haiku 4.5 (agentic) | Haiku 4.5 (one-shot) |
+| --------------------------------- | ------------- | -------------- | --------- | --------------- | ---------- | ----------- | -------------- | ------------------- | -------------------- |
+| `claude-code-20250219`            | ✓             | ✓              | ✓         | ✓               | ✓          | ✓           | ✓              | ✓                   | -                    |
+| `oauth-2025-04-20` (OAuth only)   | ✓             | ✓              | ✓         | ✓               | ✓          | ✓           | ✓              | ✓                   | ✓                    |
+| `context-management-2025-06-27`   | ✓             | ✓              | ✓         | ✓               | ✓          | ✓           | ✓              | ✓                   | -                    |
+| `prompt-caching-scope-2026-01-05` | ✓             | ✓              | ✓         | ✓               | ✓          | ✓           | ✓              | ✓                   | -                    |
+| `interleaved-thinking-2025-05-14` | ✓             | ✓              | ✓         | ✓               | ✓          | ✓           | -              | -                   | -                    |
+| `context-1m-2025-08-07`           | -             | -              | `[1m]`    | `[1m]`          | `[1m]`     | `[1m]`      | -              | -                   | -                    |
+| `effort-2025-11-24`               | -             | -              | ✓         | -               | -          | ✓           | -              | -                   | -                    |
+| `structured-outputs-2025-12-15`   | -             | `*`            | `*`       | -               | `*`        | `*`         | -              | `*`                 | `*`                  |
 
 Key rules:
 
@@ -88,7 +88,7 @@ Key rules:
 - **`context-1m` is user opt-in via `[1m]`.** Appending `[1m]` to the model string (e.g., `claude-opus-4-8[1m]`) adds the 1M beta and strips the tag before the request hits the wire. Family-based auto-enable would 400 on subscriptions or gateways that do not carry 1M access. Convention matches Claude Code.
 - **`effort` is Opus 4.6+ and Sonnet 4.6+ only.** Opus 4.5 and older, Sonnet 4.5 and older, and all Haiku variants reject it per upstream's `modelSupportsEffort`. Per-model support is encoded in `Capabilities::supported_efforts`; `accepts_effort`, `clamp_effort`, and `default_effort` keep user picks and defaults inside that set.
 - **`effort` and `context-management` betas need a body field.** Sending the header alone is a silent no-op: the request runs at the server default. See [Agentic Request Body Fields](#agentic-request-body-fields) for the matching `output_config.effort` and `context_management.edits` shapes. oxide-code pairs each capability with both its beta and its body field so the two stay in sync.
-- **`structured-outputs` is per-version and caller-opt-in.** The upstream allowlist is Opus 4.5 / 4.6+, Sonnet 4.5 / 4.6+, Haiku 4.5. The beta ships only when a caller supplies an `output_config.format` (today: the AI-title generator). The body field and header are paired on the same capability flag. A schema passed to an unsupported model silently falls back to free-form text, mirroring the `[1m]` × `context_1m` silent-strip pattern.
+- **`structured-outputs` is per-version and caller-opt-in.** The upstream allowlist is Opus 4.1+, Sonnet 4.5+, Haiku 4.5. The beta ships only when a caller supplies an `output_config.format` (today: the AI-title generator). The body field and header are paired on the same capability flag. A schema passed to an unsupported model silently falls back to free-form text, mirroring the `[1m]` × `context_1m` silent-strip pattern.
 - **Unknown model aliases** fall through substring matching on the family stem. `claude-opus-5-x` would miss every row and ship with only the identity / caching betas. Bump the `MODELS` table when a new family lands.
 
 oxide-code gates each beta header on the target model in `client::anthropic::compute_betas`, which consults the ground-truth `Capabilities` flags in `crate::model::MODELS`. New models ship by adding a row to that table, with no beta-logic changes needed.
@@ -238,7 +238,7 @@ GA as of Opus 4.6. Controls the intelligence-vs-latency tier of agentic turns vi
 
 - **The `effort-2025-11-24` beta header is necessary but not sufficient.** oxide-code used to send the header without the body field. The header became a no-op and the model ran at an undefined default.
 - **Per-model ceiling.** `max` is Opus-only, and Sonnet 4.6 400s on it. `xhigh` is Opus 4.7 / 4.8-only. `Capabilities::supported_efforts` encodes the allowed set, and `Capabilities::clamp_effort` clamps a user pick down to the highest supported level at or below it.
-- **Per-model default.** `Capabilities::default_effort` picks the highest non-`max` tier: `xhigh` on Opus 4.7 / 4.8, `high` on Opus 4.6 and Sonnet 4.6, omitted on models without effort. This tracks Claude Code for 4.6 / 4.7; oxide-code keeps `xhigh` as the 4.8 default even though upstream lowered the stock 4.8 default to `high`.
+- **Per-model default.** The raw API default is `high` for every effort-capable model (`high` is equivalent to omitting the field), and `xhigh` is Anthropic's coding recommendation rather than a stock default. Claude Code bumps Opus 4.7 to `xhigh` but rides the `high` default on Opus 4.8; oxide-code mirrors this in `Capabilities::default_effort` (`xhigh` on 4.7, `high` on 4.8 / 4.6 / Sonnet 4.6, omitted on models without effort).
 - **`max_tokens` should scale with effort.** Claude Code uses 64 K on Opus 4.7 at `xhigh`, 32 K on Sonnet 4.6 at `high`. oxide-code's `default_max_tokens(effort)` matches the upper tiers and uses 16 K otherwise when the user hasn't set `ANTHROPIC_MAX_TOKENS` explicitly.
 
 ### `context_management.edits`
diff --git a/docs/research/tui/status-line.md b/docs/research/tui/status-line.md
index 568b91d8..f71c8f8f 100644
--- a/docs/research/tui/status-line.md
+++ b/docs/research/tui/status-line.md
@@ -39,11 +39,12 @@ opencode keeps usage in the prompt footer rather than a fully configurable statu
 
 Anthropic's pricing page lists first-party Claude API prices in USD per million tokens and separate prompt-cache rates for 5-minute writes, 1-hour writes, cache reads, and output tokens. Checked on 2026-05-14, the relevant rows for oxide-code's model table are:
 
-| Family                     | Input | 5m cache write | 1h cache write | Cache read | Output |
-| -------------------------- | ----- | -------------- | -------------- | ---------- | ------ |
-| Opus 4.8 / 4.7 / 4.6 / 4.5 | $5    | $6.25          | $10            | $0.50      | $25    |
-| Sonnet 4.x                 | $3    | $3.75          | $6             | $0.30      | $15    |
-| Haiku 4.5                  | $1    | $1.25          | $2             | $0.10      | $5     |
+| Family     | Input | 5m cache write | 1h cache write | Cache read | Output |
+| ---------- | ----- | -------------- | -------------- | ---------- | ------ |
+| Opus 4.5+  | $5    | $6.25          | $10            | $0.50      | $25    |
+| Opus 4.1   | $15   | $18.75         | $30            | $1.50      | $75    |
+| Sonnet 4.x | $3    | $3.75          | $6             | $0.30      | $15    |
+| Haiku 4.5  | $1    | $1.25          | $2             | $0.10      | $5     |
 
 Cost display should stay best-effort because account discounts, marketplace billing, data residency, fast mode, and server-side tool pricing can change the final bill.