Mistral Large 4 API Pricing: $1.36 In, $4.18 Out, in Public Preview

Short answer. Mistral Large 4 costs $1.36 per million input tokens, $0.14 for cached input and $4.18 for output, on Mistral’s pricing page on October 22, 2026. That is 2.72 times Mistral Large 3’s input price and 2.79 times its output price. Large 4 has been in Public Preview since October 6, 2026; Large 3 is still generally available.

In plain words. Mistral has a new large model, Large 4. Each request costs almost three times what it costs on Large 3, and Large 4 reads much longer documents. It is still a preview: Mistral may change it, and may retire it before it is final.

Update, October 30, 2026: Mistral has cut Mistral Large 4’s price. On October 30, 2026 its pricing page printed Mistral Large 4 ↗ | $0.7 | $0.07 | $2.1 |: $0.70 per million input tokens, $0.07 for cached input and $2.10 for output, down from the $1.36, $0.14 and $4.18 quoted in this post. Mistral does not say when the price changed. The model is still in Public Preview. The figures below are the ones read on October 22, 2026; the change is recorded on the price changes page and explained in Mistral Large 4’s price cut.

Mistral Large 4 is the newest model in the “Flagship models” table of Mistral’s API pricing page. Its model page describes it in one sentence: “Mistral Large 4 is a state-of-the-art, open-weight, general-purpose multimodal model with a granular Mixture-of-Experts architecture.” It sits at the top of the pricing page’s “Flagship models” table, above Mistral Large 3, which Mistral still sells at its old price. This post sets the two side by side, works out what one request costs on each, and quotes what Mistral says a Public Preview model is.

The prices, side by side

Mistral’s pricing page has a table headed “Model | Input | Cached input | Output |”, in US dollars per million tokens. On October 22, 2026 its first two rows read “Mistral Large 4 ↗ | $1.36 | $0.14 | $4.18 |” and “Mistral Large 3 ↗ | $0.5 | $0.05 | $1.5 |”. The model pages give the context window: “Context 1M” for Large 4 and “Context 256k” for Large 3.

So every price of Large 4 is between 2.72 and 2.8 times the matching price of Large 3, and the context window is about four times as large. The cached input price is the price of input tokens the API has already seen in an earlier request and keeps in its cache.

What one request costs on each

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

• Mistral Large 4: 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.
• Mistral Large 3: 10,000 × $0.50 ÷ 1,000,000 = $0.005 for input, plus 1,000 × $1.50 ÷ 1,000,000 = $0.0015 for output, so $0.0065.

A thousand such requests cost $17.78 on Large 4 and $6.50 on Large 3, which is 2.74 times as much. A prompt longer than 256K tokens has no Large 3 figure at all: it does not fit in Large 3’s context window.

What “Public Preview” means at Mistral

Large 4’s model page says “October 6, 2026 Public Preview Open v 26.10”. Large 3’s says “December 2, 2025 Blog GA Apache 2.0 v 25.12”. Mistral’s model lifecycle page explains the difference. Its table headed “Stage | Description | Silent updates |” has the rows “Public Preview | Near-final before General Availability. | Yes |” and “General Availability | Production-ready and stable. | No |”.

The same page lists what a Public Preview model is. It says such models “Are priced at the same rate as General Availability models.”, “Allow silent updates.” and “Have no guaranteed path to General Availability, and may be retired before reaching it.”

The notice before a model is withdrawn is also shorter. The deprecation table reads “Public Preview | 1 month |” and “General Availability | 6 months |”, and the page says “Deprecation is announced as soon as a replacement model is available. The replacement can be in General Availability or Public Preview.” For anyone building on Large 4 today, that means a month’s notice, not six.

Mistral does not say Large 4 replaces Large 3

Mistral’s pages do not call Large 4 a replacement for Large 3, and Large 3 has no deprecation notice. Large 4’s page lists two identifiers, mistral-large-4 and mistral-large-4-0. Large 3’s page lists mistral-large-2512 and mistral-large-latest. The lifecycle page describes that last kind of alias as “model-name-latest | mistral-medium-latest | Latest version, across generations |”, and 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.”

So on October 22, 2026, a request to mistral-large-latest still went to Large 3, at $0.50 and $1.50. If Mistral moves that alias to Large 4, the same code would pay $1.36 and $4.18 without a change on the caller’s side. Pinning mistral-large-2512 or mistral-large-4-0 avoids that.

When we saw it

Mistral’s model page dates Large 4 October 6, 2026. This site saw it on Mistral’s pricing page on October 21, 2026, through its daily scan of that page, and added it on October 22, 2026.

While reading the page we also found that seven Mistral rows lacked the cached input price the page prints (Mistral Large 3 at $0.05, Mistral Medium 3.5 at $0.15, Mistral Small 4 at $0.015, Ministral 3 14B at $0.02, Ministral 3 8B at $0.015, Ministral 3 3B at $0.01 and Codestral at $0.03), and added them on October 22, 2026 as a correction of our own, listed on price changes.

Le Chat and the Mistral plans

Mistral’s plans page for its chat app, read on October 22, 2026, does not mention Large 4. Its only mention of Mistral Large is an API example at Large 3’s prices: “For example, Mistral Large costs $0.5 /M tokens in and $1.5 /M tokens out.” The plans and their prices are on the Mistral subscription page.

Both models have their own page with every figure and its source: Mistral Large 4 and Mistral Large 3. Every other Mistral model, and those of the other providers, are in the API pricing table.

How we checked

Mistral’s pricing page, the model pages of Large 4 and Large 3, the models overview, the model lifecycle page and Mistral’s plans page were read by a script from a server in Finland on October 22, 2026. The pricing page showed Standard prices in US dollars. Large 4 first appeared in the daily scan of the pricing page on October 21, 2026; the scan only flags new models and never writes prices. Each price and line quoted here comes from the saved text of the October 22 readings, and a test checks every quote and every figure against it. The multipliers and the cost of the sample request are worked out from those prices, with the sums shown.

Frequently asked questions

How much does Mistral Large 4 cost?

Mistral's pricing page, read on October 22, 2026, lists Mistral Large 4 at $1.36 per million input tokens, $0.14 for cached input and $4.18 for output. Mistral Large 3 costs $0.50, $0.05 and $1.50. (Update, October 30, 2026: Mistral's pricing page, read that day, lists $0.70 input, $0.07 cached input and $2.10 output per million tokens.)

Is Mistral Large 4 generally available?

No: its model page says 'October 6, 2026 Public Preview Open v 26.10'. Mistral's lifecycle policy says Public Preview models 'Have no guaranteed path to General Availability, and may be retired before reaching it.'

Does Mistral Large 4 replace Mistral Large 3?

Mistral does not say so: on October 22, 2026, Mistral Large 3 was still marked GA and its page listed the alias mistral-large-latest. Large 4's page lists only mistral-large-4 and mistral-large-4-0.

Is Mistral Large 4 in Le Chat?

Mistral's plans page, read on October 22, 2026, does not mention Mistral Large 4. Its only mention of Mistral Large is an API example at Large 3's prices.