Skip to content

alibaba-cn/qwen-plus: add published explicit cache rates - #5235

Open
tbourrillon wants to merge 1 commit into
anomalyco:devfrom
tbourrillon:qwen-plus-cache-rates
Open

alibaba-cn/qwen-plus: add published explicit cache rates#5235
tbourrillon wants to merge 1 commit into
anomalyco:devfrom
tbourrillon:qwen-plus-cache-rates

Conversation

@tbourrillon

Copy link
Copy Markdown

alibaba-cn/qwen-plus carries input, output and reasoning but no cache rates, even though Alibaba publishes explicit context-cache pricing for the model. Anything pricing a cached qwen-plus turn currently has to fall back to the uncached input rate.

From the Model Studio pricing page (China Beijing tab; Germany Frankfurt and US Virginia show the same figures under Scope: Global), base tier Input <= 128k:

Billing item USD / 1M tokens
Explicit Cache Creation 0.144
Explicit Cache Read 0.012

Both check out against Alibaba's documented multipliers and the input rate this entry already carries. The context cache guide states explicit cache creation is billed at 125% of the standard input price and an explicit cache hit at 10%:

  • 0.115 × 1.25 = 0.14375 → published 0.144
  • 0.115 × 0.10 = 0.0115 → published 0.012

So two independent sources agree, against a base price this file already had right.

On which rate cache_read carries

I used the explicit hit rate (0.012), which is how the other Qwen entries here are populated — qwen3.7-flash, qwen3.6-plus and qwen3.7-plus all sit at exactly 10% of their input.

Worth flagging that Alibaba prices two different kinds of hit: an explicit (cache_control) hit at 10%, and an implicit prefix-cache hit at 20% — 0.023 for this model, also published on that page. There's one cache_read field, and the API reports both kinds in the same cached_tokens, so a consumer can't tell which rate applied. I've left that alone here rather than fold a schema question into a data fix; happy to open it separately if that's useful.

Not included

qwen-plus is also published in three context tiers (<= 128k, 128k–256k, 256k–1M) and this entry carries only the first, flat, on a model whose limit.context is 1,000,000. I have figures for the upper tiers but only from a rendered read of the page, and they contain an internal inconsistency I couldn't reconcile confidently, so I've deliberately left them out rather than guess. Raising that separately so someone with a cleaner source can do it properly.


Verified against the live page on 2026-08-21. Happy to adjust the format or split this differently.

Alibaba publishes explicit context-cache pricing for qwen-plus that this
entry does not carry, so anything pricing a cached qwen-plus turn has to
fall back to the uncached input rate.

From the Model Studio pricing page (China Beijing / Global scope), base
tier (Input <= 128k):

  Explicit Cache Creation  0.144 per 1M tokens
  Explicit Cache Read      0.012 per 1M tokens

Both corroborate Alibaba's documented multipliers against the input rate
this entry already carries: the context-cache guide states explicit cache
creation is billed at 125% of the standard input price and an explicit
cache hit at 10%. 0.115 * 1.25 = 0.14375 (published 0.144) and
0.115 * 0.10 = 0.0115 (published 0.012).

cache_read is the EXPLICIT hit rate, matching how the other Qwen entries
in this repo are populated. Alibaba prices an implicit prefix-cache hit
separately at 20% (0.023), which this schema has no field for; raised as
an issue rather than folded in here.
@github-actions

Copy link
Copy Markdown
Contributor

No actionable findings.

@github-actions github-actions Bot added the reviewer: ready Automated review found no actionable items label Aug 21, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

reviewer: ready Automated review found no actionable items

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant