Mistral Large 4 Price Cut: $0.70 In, $2.10 Out, Still in Public Preview

Short answer. Mistral has cut Mistral Large 4’s price. On October 30, 2026 its pricing page lists $0.70 per million input tokens, $0.07 for cached input and $2.10 for output, down from $1.36, $0.14 and $4.18 on October 26. A request of 10,000 input and 1,000 output tokens now costs $0.0091 instead of $0.01778. The model is still in Public Preview.

In plain words. Mistral’s newest large model now costs about half of what it cost a few days ago. It is still a preview, so Mistral may change it again. Mistral has not said when the price changed, or whether the new price is for a limited time.

Mistral Large 4 entered Mistral’s API in Public Preview on October 6, 2026. When AI Signal first read its price, on October 22, it was the second dearest of Mistral’s own models on its pricing page, after Mistral Medium 3.5, at 2.72 times Mistral Large 3’s input price (the post of that day). Eight days later the same row of the same table prints about half of every figure. This post is about that change: the figures before and after, what a request costs now, and where the new price sits next to Mistral’s other models and to models of other makers in AI Signal’s table.

Before and after

Mistral’s pricing page has a table headed “Model | Input | Cached input | Output |”, in US dollars per million tokens. AI Signal’s price check of October 26, 2026 read Large 4’s row as “Mistral Large 4 ↗ | $1.36 | $0.14 | $4.18 |”. The check of October 30, 2026 read “Mistral Large 4 ↗ | $0.7 | $0.07 | $2.1 |”. The Markdown version of the model page says the same on October 30: “Pricing: Input $0.7 /M Tokens; Cached input $0.07 /M Tokens; Output $2.1 /M Tokens”.

Input and output did not fall by exactly the same factor: the new input price is 51% of the old one, the new output price 50%. The cached input price is still a tenth of the input price, as it was before. The context window did not change: the model page still says “Context 1M”.

What one request costs now

Take a request with 10,000 input tokens and 1,000 output tokens, with no cache:

• Before: 10,000 × $1.36 ÷ 1,000,000 = $0.0136 for input, plus 1,000 × $4.18 ÷ 1,000,000 = $0.00418 for output, so $0.01778.
• Now: 10,000 × $0.70 ÷ 1,000,000 = $0.007 for input, plus 1,000 × $2.10 ÷ 1,000,000 = $0.0021 for output, so $0.0091.

A thousand such requests cost $9.10 now and cost $17.78 at the old price, 1.95 times as much. A request that reuses a cached prompt pays $0.07 instead of $0.70 per million for the cached part.

Next to Mistral’s other models

The same table, read on October 30, 2026, still prints “Mistral Large 3 ↗ | $0.5 | $0.05 | $1.5 |” and “Mistral Medium 3.5 ↗ | $1.5 | $0.15 | $7.5 |”.

On October 22, Large 4 cost 2.72 times Large 3’s input price and 2.79 times its output price. Now both ratios are 1.4: $0.70 ÷ $0.50 and $2.10 ÷ $1.50. The sample request costs 1.4 times as much on Large 4 as on Large 3 ($0.0091 ÷ $0.0065), and Medium 3.5’s costs 2.47 times as much as Large 4’s ($0.0225 ÷ $0.0091). Large 4 is still the second dearest of Mistral’s own models, after Medium 3.5; the gap to Large 3 is what shrank. The models Mistral serves for other makers, such as Z.ai’s GLM 5.3 and DeepSeek V4.1 Flash 1017, are listed apart on Mistral’s API pricing page here.

Next to models of other makers at a similar price

AI Signal’s table has rows from other makers with a context window of one million tokens or more. These are the five whose sample request (10,000 in, 1,000 out) costs closest to Large 4’s $0.0091, with the price each row stores and the day it was last verified:

Three conditions sit behind those figures. Gemini 3.6 and 3.8 Flash are at a promotional price through December 31, 2026; from January 1, 2027 Google’s page lists $1.50 and $7.50, which would put their sample request at $0.0225. Qwen3.7-Plus is Alibaba’s Singapore price, and a prompt over 256,000 tokens pays its second price, $1.20 and $4.80. Large 4 has one price for its whole context window on Mistral’s page. The comparison is by price only: it says nothing about which model does a given job better. The full ranking is in the API pricing table.

Still a preview, and the mistral-large-latest alias

The Markdown version of Large 4’s model page, read on October 30, 2026, says “Status: Public Preview”, and the page itself says “October 6, 2026 Public Preview Mistral Ecosystem Open Weight License v 26.10”. On October 22 the same line read “October 6, 2026 Public Preview Open v 26.10”: the licence name is new, the stage is not.

Mistral’s model lifecycle page says Public Preview models “Are priced at the same rate as General Availability models.” and “Allow silent updates.”, and gives them a deprecation notice of “Public Preview | 1 month |”.

The alias did not move. Large 3’s model page, read on October 30, 2026, lists “API names: mistral-large-2512, mistral-large-latest” and “Status: GA”. Large 4’s lists “API names: mistral-large-4, mistral-large-4-0”. The lifecycle page warns: “Using an alias may expose you to silent updates in model behavior and pricing. For precise control, pin your deployment to a specific major.minor version identifier.” A request to mistral-large-latest therefore still goes to Large 3, at $0.50 and $1.50.

What Mistral says about the change

Nothing that gives a date. The pricing page and the model page print the new figures and no note. Mistral’s changelog, read on October 30, 2026, has no entry about Large 4 after its launch. Under October 6 it still prints the launch entry, which ends with one line about price: “Launch pricing: 50% off for 2 weeks.” AI Signal first read Large 4’s price on October 22, sixteen days after that entry, at $1.36 and $4.18, and read the same figures again on October 26. No page says whether the price of October 30 is a new list price or a limited one.

Because the day of the change is not known, AI Signal records it on the price changes page with the first day the new price was seen, October 30, 2026, and quotes the last reading at the old price.

How we checked

• Prices: Mistral’s pricing page (docs.mistral.ai/inference/pricing), read by script on October 30, 2026, table “Model | Input | Cached input | Output |”. The old figures are AI Signal’s own readings of the same table on October 22 and October 26, 2026.
• Model pages: Mistral Large 4 (docs.mistral.ai/models/mistral-large-4-0), Mistral Large 3 (docs.mistral.ai/models/mistral-large-3-25-12) and the Markdown version of Mistral Medium 3.5’s (mistral-medium-3-5-26-04.md, “Status: GA”), read on October 30, 2026.
• Lifecycle and changelog: docs.mistral.ai/inference/model-lifecycle and docs.mistral.ai/getting-started/changelog, read on October 30, 2026.
• Other makers: the five rows of AI Signal’s table quoted above, each with the date it was last verified on its maker’s page.
• Arithmetic: the sample request and every ratio are computed here from the prices quoted, and rounded only in the text.

The Mistral Large 4 page carries the current price and the dated change.

Frequently asked questions

How much does Mistral Large 4 cost now?

Mistral's pricing page, read on October 30, 2026, lists Mistral Large 4 at $0.70 per million input tokens, $0.07 for cached input and $2.10 for output. On October 26, 2026 it listed $1.36, $0.14 and $4.18.

When did Mistral cut the price of Mistral Large 4?

Mistral does not say. The old price was last read on October 26, 2026 and the new one first on October 30, 2026; neither the pricing page, the model page nor the changelog gives a date for the change.

Is the new Mistral Large 4 price temporary?

No page of Mistral's says so. Its changelog still prints, under October 6, 'Launch pricing: 50% off for 2 weeks.', and has no later entry about Mistral Large 4.

Does mistral-large-latest point to Mistral Large 4?

No: on October 30, 2026, Mistral Large 3's page lists the API names mistral-large-2512 and mistral-large-latest and the status GA. Large 4's page lists mistral-large-4 and mistral-large-4-0 and the status Public Preview.