feat: add token market provider

[FitBBC]

Add Token Market Provider

This PR adds Token Market as an OpenAI-compatible model provider.

Provider

• OpenAI-compatible API
• Supports streaming
• Supports tool calling
• Supports reasoning models
• Provides access to models from Anthropic, OpenAI, Google,
  DeepSeek, Qwen, GLM, MiniMax, and others.

API

Base URL:

https://api.tokensmarket.ai/v1

Authentication:

Authorization: Bearer <API_KEY>

Changes

• Add Token Market provider metadata
• Add Token Market logo
• Add supported model mappings and pricing

Data Sources

The provider metadata in this PR is based on Token Market's official API
and documentation.

Provider and model availability

• Token Market API:
  GET https://api.tokensmarket.ai/v1/models

This endpoint is used to verify that the submitted model IDs are currently
available through Token Market.

Pricing

• Token Market pricing/catalog:
  https://www.tokensmarket.ai/models

The cost.input, cost.output, cost.cache_read, and cost.cache_write
values in the provider overrides are taken from Token Market's current
published pricing/catalog data.

All prices in models.dev are expressed in USD per million tokens unless the
underlying model uses a different billing unit.

Model capabilities and modalities

For models using base_model, general model metadata such as context limits,
tool-calling support, reasoning support, and modalities is inherited from the
corresponding canonical models/<lab>/... entry.

Provider-specific overrides are included only where Token Market exposes a
different supported surface.

Token Market model availability and provider-specific modality support were
verified against:

• GET https://api.tokensmarket.ai/v1/models
• Token Market documentation:
  https://www.tokensmarket.ai/docs

Provider-specific behavior

Reasoning controls, caching behavior, and other provider-specific overrides
are only declared where they are supported by Token Market's documented API
behavior.

Relevant API documentation:

• OpenAI-compatible API:
  https://www.tokensmarket.ai/docs
• Reasoning:
  https://www.tokensmarket.ai/docs
• Prompt caching:
  https://www.tokensmarket.ai/docs

[github-actions]

Action items

• [critical] [violation] providers/tokensmarket/logo.svg:1 - Check: New provider logo.svg must use currentColor, have no fixed width/height, and preferably a square viewBox. Why: The SVG sets width="24" height="24", uses a non-square viewBox="0 0 416 480", and the path has no fill="currentColor" (default fill is opaque black), so the logo will not theme correctly. Action: Rewrite the logo as a size-free SVG with fill="currentColor" (or equivalent strokes) and a square viewBox (e.g. 0 0 24 24).
• [high] [violation] providers/tokensmarket/models/seedance-2-0.toml:8 - Check: Non-lab hosts must use base_model; if lab metadata is missing, add a complete models/<lab>/<id>.toml first. Why: Token Market did not create Seedance/Seedream/HappyHorse/Qwen embedding-rerank models, but these entries are full inline definitions with no base_model, and no lab files exist under models/. Same pattern on seedance-2-0-fast, seedance-2-0-mini, seedance-2-5, seedream-4-0, seedream-4-5, seedream-5-0-lite, seedream-5.0-pro, happyhorse-1.0-video-edit, happyhorse-1.1-*, qwen3-embedding-8b, qwen3-rerank, and qwen3.7-text-embedding. Action: Add complete lab metadata under the correct labs (e.g. ByteDance Seed / Alibaba), then convert these provider files to override-only base_model entries (cost/status/host deltas only).
• [high] [violation] providers/tokensmarket/models/mimo-v2.5-pro.toml:6 - Check: Relay reasoning_options must match the lab / same-surface peer control set, not an invented effort ladder. Why: First-party providers/xiaomi/models/mimo-v2.5-pro.toml and established peers expose only toggle, but this file authors effort none|low|medium|high. That invents graded levels the lab surface does not document. Action: Align with lab/peers (toggle only, with a leading wire-path comment), or provide host-specific evidence that this gateway actually accepts and applies that full effort enum for MiMo V2.5 Pro.
• [medium] [violation] providers/tokensmarket/models/claude-haiku-4-5.toml:5 - Check: After base_model, keep only provider-specific fields and real deltas. Why: description is restated identically to models/anthropic/claude-haiku-4-5.toml / first-party Anthropic. Action: Remove the redundant description override.
• [medium] [violation] providers/tokensmarket/models/kimi-k2.7-code.toml:5 - Check: Override-only provider files must not restate unchanged lab fields. Why: family = "kimi-k2" already exists on models/moonshotai/kimi-k2.7-code.toml. Action: Drop the redundant family key.
• [medium] [possible mistake] providers/tokensmarket/models/hy3.toml:6 - Check: Effort values should match native/peer surfaces for the model. Why: This sets low|medium|high, while established relays (e.g. Vercel/OpenRouter peers) use none|low|high and map off via none/no_think. Including medium and omitting none may misstate the reachable control surface. Action: Verify Token Market’s Hy3 wire values against Tencent/upstream peers and correct the effort list (likely include none and drop unverified medium).
• [low] [possible mistake] .pr-review/pull-request.json:4 - Check: Data-changing PRs should cite direct pricing/docs and map each citation to claims. Why: The body links docs and the models API, but does not map sources to the many USD costs, modality cuts, or per-request/video pricing omissions. Action: Add first-party Token Market (or upstream vendor) pricing/docs citations and state what each URL supports.

[github-actions]

Action items

• [high] [violation] providers/tokensmarket/logo.svg:1 - Check: New provider logos must use currentColor for fills/strokes with no hardcoded colors. Why: The SVG has a bare <path> and no fill="currentColor" (or equivalent), so the mark paints as the default black fill and will not follow theme/current color. Action: Set fill="currentColor" on the root <svg> or path(s) and ensure there are no hardcoded colors.
• [high] [possible mistake] providers/tokensmarket/models/deepseek-v4-flash.toml:17 - Check: Relay reasoning options must match the lab/same-surface peer set for this host path. Why: This file is documented as the Alibaba/DashScope channel, but effort includes low. DashScope peers (providers/alibaba/models/deepseek-v4-flash-0731.toml, providers/alibaba-cn/models/deepseek-v4-flash.toml) expose only high/max because low/medium collapse to high. Action: Align to ["high", "max"] for the DashScope surface, or provide host evidence that Token Market preserves a distinct low.
• [medium] [possible mistake] providers/tokensmarket/models/kimi-k2.7-code.toml:5 - Check: Do not claim toggle/budget unless this host actually exposes those controls. Why: First-party and major peers use always-on reasoning_options = [] (moonshotai, OpenRouter, opencode). The closest DashScope peer notes thinking cannot be disabled, yet this entry authors toggle + budget_tokens. Action: Verify Token Market can turn thinking off and accept a reasoning budget for this ID; if not, use [] (or budget-only if only the budget field is real).
• [medium] [possible mistake] providers/tokensmarket/models/longcat-2.0.toml:6 - Check: budget_tokens only when this host has a real reasoning-budget field for the model. Why: First-party LongCat is toggle-only (providers/longcat/models/LongCat-2.0.toml). The file itself says upstream exposes only enabled/disabled thinking, but still adds budget_tokens with bounds. Action: Drop budget_tokens unless Token Market/SiliconFlow docs or a live request show a meaningful LongCat reasoning budget on this path.
• [medium] [violation] providers/tokensmarket/models/claude-sonnet-5.toml:8 - Check: Every toggle needs a leading top-of-file comment with the exact wire path. Why: This model authors toggle + effort, but the header only says Anthropic Messages passthrough and does not document the on/off field (e.g. thinking.type = enabled|disabled). Action: Add an exact toggle wire comment at the top of the file.
• [medium] [violation] providers/tokensmarket/models/mimo-v2.5-pro.toml:6 - Check: Every toggle needs a leading top-of-file comment with the exact wire path. Why: reasoning_options is a plain toggle, but the comment only says thinking can be turned on or off with no request field/path. Action: Document the exact Token Market/upstream toggle field syntax in the leading comment.
• [low] [possible mistake] .pr-review/pull-request.json:4 - Check: Data-changing PRs should cite direct pricing/docs mapped to claims. Why: The body names the API/docs base URL but does not map sources to the many non-round USD/MTok rates, cache prices, or modality overrides. Action: Add first-party Token Market catalog/pricing citations and state what each supports (list prices, cache rates, modality limits).

[github-actions]

Checking a few more peer files and the logo final form to lock findings.