Skip to content

[mistral-ai] Update -latest aliases to Large 4 / Small 4 and add their dated keys - #1039

Open
Mattbusel wants to merge 1 commit into
Portkey-AI:mainfrom
Mattbusel:fix/mistral-large-4-small-4-pricing
Open

Mattbusel wants to merge 1 commit into
Portkey-AI:mainfrom
Mattbusel:fix/mistral-large-4-small-4-pricing

Conversation

@Mattbusel

@Mattbusel Mattbusel commented Nov 18, 2026 •

Copy link
Copy Markdown

Description

mistral-large-latest and mistral-small-latest are still priced as Large 3 and Small 3.2, but Mistral moved both aliases to the new models, so calls through them are billed low right now.

From Mistral's model pages and pricing page (checked 2026-11-17), USD per 1M tokens:

key now serves in / cached in / out stored before
mistral-large-latest Mistral Large 4 (26.10, GA 2026-11-05) 0.70 / 0.07 / 2.10 0.50 / none / 1.50
mistral-small-latest Mistral Small 4 (26.03) 0.15 / none listed / 0.60 0.10 / none / 0.30

So Large is under-billed about 29% on both input and output, and Small output is under-billed by half.

What changed:

  • pricing/mistral-ai.json: repointed the two -latest rows to the prices above. Added mistral-large-4 and mistral-large-4-0 (the dated API names on the Large 4 model page) and mistral-small-2603 (the dated name for Small 4). The Large 4 rows get cache_read_input_token at the published cached-input price and cache_write_input_token at 0, same shape as deepseek.json.
  • general/mistral-ai.json: added the three new keys as chat with tools and image (both models take text and image input and support function calling per their model pages).
  • Values are cents per token, e.g. $0.70/1M = 0.00007, $0.07/1M = 0.000007.

What I left alone, on purpose:

  • mistral-large-2512 stays at 0.50 / 1.50. That is still the right Large 3 price; Mistral lists it as deprecated (retiring 2027-04-30), not repriced.

  • Small 4 and the Ministral 3 rows get no cached-input price, because Mistral does not publish one for them. Only Large 4 has a cached-input line on the pricing page.

  • The default.calculate block in this file only multiplies input_tokens by request_token, so the new cache_read_input_token value is there for when cached tokens are counted, same as the DeepSeek file today. If you want it to bill now, I can add the cache_read_tokens term to the Mistral calculate block in this PR or a follow-up, matching openai.json.

  • New model pricing

  • Update existing pricing

  • New model configuration

  • Bug fix

Source Verification

Source Link: https://docs.mistral.ai/inference/pricing (aliases and dated names: https://docs.mistral.ai/models/mistral-large-4-0 and https://docs.mistral.ai/models/mistral-small-4-0-26-03)

Checklist

  • I have validated the JSON using jq or an online validator (jq empty on every file in pricing/ and general/, plus python3 scripts/check_duplicate_keys.py, both clean)
  • I have verified that prices are in cents per token (not dollars)
  • I have included the source link above
  • I have signed the CLA (if first-time contributor)

Related Issue

No issue filed for this one. I found it with an automated cross-check of this repo's prices against llm-prices (https://github.com/Mattbusel/llm-prices), a provider price table I keep checked against vendor pages, then confirmed every number by hand on Mistral's pages before writing it. If it is useful, I can turn that cross-check into a small GitHub Action that opens a diff here when a provider's published price moves, so the next alias switch shows up the week it happens.

Disclosure: I maintain llm-prices (https://github.com/Mattbusel/llm-prices).

… dated keys

mistral-large-latest now serves Mistral Large 4 (26.10) at $0.70 in,
$0.07 cached in, $2.10 out per 1M. mistral-small-latest serves
Mistral Small 4 (26.03) at $0.15 in, $0.60 out per 1M. Adds
mistral-large-4, mistral-large-4-0 and mistral-small-2603 to pricing
and general configs.

Source: https://docs.mistral.ai/inference/pricing

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant