Mistral Large 4 API Pricing (Verified Oct 6, 2026)

Mistral Large 4 price per million tokens

Right now you pay half the list price. Mistral’s pricing page shows $1.36 / $4.18 struck through and labels $0.68 / $2.09 as a temporary sale price. The discount is exactly 50% on every line: 0.68 ÷ 1.36, 2.09 ÷ 4.18 and 0.07 ÷ 0.14 all equal 0.50.

Cached input is the cheapest line on the bill. A cached token costs about one tenth of a fresh input token ($0.07 versus $0.68 at the sale price), so long system prompts and repeated documents get much cheaper on the second call.

Batch, priority and regional tiers

Batch is the cheapest way to run Mistral Large 4: it takes another 50% off, so at the current sale price a million input tokens costs $0.34 and a million output tokens costs $1.045. Priority processing adds 75% for faster, reserved capacity, and regional inference (for example the Mistral-operated European deployment) adds 10%.

All four tiers carry the same temporary 50% sale, so the gap between them stays the same whichever price applies when you read this.

Why OpenRouter and Vercel show $0.68 / $2.09

OpenRouter and Vercel AI Gateway pass through Mistral’s own sale price, so the per-token rate is the same on all three. OpenRouter’s endpoint data for mistralai/mistral-large-4-0 carries a discount value of 0.5, and Mistral is the only provider serving the model there.

Mistral has not announced when the sale ends. Some summaries claim the discount runs for the first two weeks, but no Mistral page states a date. Benchmark sites such as Artificial Analysis and Vals AI calculate their cost figures at the $1.36 / $4.18 list price, so budget at list price if you are planning beyond this month.

What real workloads cost

Output tokens drive most bills, because they cost about three times as much as input. The examples below use the formula tokens ÷ 1,000,000 × price and assume 30 days a month; plug your own numbers into the calculator above for a closer estimate.

Reasoning makes output longer. With reasoning_effort set to "high", the model returns thinking chunks before its answer, and Artificial Analysis recorded 200M output tokens for its full test run against a median of 81M across models, so it rates Mistral Large 4 as very verbose. Setting reasoning_effort to "none" for simple tasks keeps output short.

Mistral Large 4 price vs DeepSeek V4, GPT-6, Claude and others

At the sale price, Mistral Large 4 costs about the same as DeepSeek V4 Pro at off-peak hours and far less than the US flagship models. At list price it is 2.9× cheaper on input and 4.8× cheaper on output than Claude Opus 5.5, and 7.4× / 12× cheaper than GPT-6 Astra. Inside Mistral’s own lineup it costs more than Mistral Large 3 ($0.50 / $1.50) but less per output token than Mistral Medium 3.5 ($7.50).

The last column applies one fixed workload (10M input + 2M output tokens) to every model, which makes the gap easy to read. For score-for-score comparisons, see the benchmarks page and the /vs-deepseek-v4 and /vs-mistral-large-3 guides.

Cost per benchmark task

Independent testers put a price on whole tasks, not just tokens. Artificial Analysis reports $1.13 per Intelligence Index task for Mistral Large 4 at list price, and Vals AI reports $13.78 per test on its Vals Index. Both include the extra output from reasoning, so they are a realistic upper bound for hard, multi-step work.

Five ways to cut your Mistral Large 4 bill

The biggest savings come from batch jobs and caching; the rest come from controlling output length.

• Send offline jobs (evaluations, bulk classification, document backfills) to the Batch endpoint /v1/batch for 50% off every token type.
• Keep long system prompts and shared documents identical at the start of each request so they can be billed at the cached rate of $0.07 instead of $0.68 per 1M.
• Use reasoning_effort "none" for lookups, rewrites and extraction; keep "high" for code, math and multi-step agents where the thinking pays off.
• Cap max_tokens on endpoints that only need short answers, because each 1M output tokens costs $2.09 at the sale price.
• Route simple traffic to Mistral Small 4 ($0.15 / $0.60) and keep Mistral Large 4 for the requests that need its coding, legal and vision strength.

Prefer a flat monthly price?

If you want to use Mistral Large 4 without an API key or token math, chat with it on our homepage. Anyone gets 3 free messages a day without an account, a free account gets 15 credits a day, and Pro costs $39.90 a month or $199.90 a year for 3,000 credits a month, where one credit is about one short reply. Plan details are on the /pricing page, and the /free guide lists every free way to try the model.