PUBLIC PREVIEWLAUNCH SALE -50%1M CONTEXTTEXT + VISIONWEIGHTS PROMISED1T-PARAM MOE
Mistral Large 4 API Pricing
Mistral Large 4 is Mistral's largest model to date, released as a public preview on October 6, 2026: a natively multimodal Mixture-of-Experts with 52B active and about 1T total parameters and a 1M-token window. The list price is $1.36/M input, $4.18/M output and $0.14/M cached input, but a launch sale charges half that - $0.68 / $2.09 / $0.07 - and Mistral's changelog says it runs for two weeks. Mistral calls it open-weight; the weights are promised by the end of October and are not out yet. Checked by hand against docs.mistral.ai.
Input - per 1M tokens
$1.36/M
Sale $0.68 - 50% off list price
Output - per 1M tokens
$4.18/M
Sale $2.09 - 50% off list price
Cached input
$0.14/M
Cache 10% of input, per model -90%
Effective - agentic blend
$0.665/M
92/8 split - 82% cache, list
§ 01 / TERMINAL
Run the numbers.
Calculator pre-loaded with Mistral Large 4's list rates. While the launch sale runs, Mistral charges half of every figure here; the calculator shows the list price because that is what the model costs once the sale ends.
$ /mo
Workload split
Prompt cache hit rate
Tokens you can process
—
Words equivalent (English)
—
Effective rate
—
§ 02 / SCENARIOS
Real-world presets.
CODING AGENT
Repo-wide bug fix
$0.093/task
LONG DOC
Contract risk review
$0.120/doc
VISION
Engineering drawing check
$0.019/drawing
SECURITY OPS
Vulnerability triage
$0.038/finding
Calibrated · measured on the vendor's tokenizer · 2026-06-10 Auto-counts as you type
Counts use a chars-per-token calibration measured on the vendor's own published tokenizer (mistralai/Mistral-Nemo-Instruct-2407, 2026-06-10). English prose is typically within a few percent; code and non-Latin scripts tokenize heavier. For billing-exact counts use the vendor's count-tokens API.
Characters 537
Words 90
Tokens (estimated) 106 tokens
Cost as input · uncached $0.00014 USD
Cost as output · uncached $0.00044 USD
Cost as cached input $0.00001 USD
| Model | Input /M | Output /M | Effective blended | Context | Best for |
|---|---|---|---|---|---|
| Mistral Large 4 Current | $1.36 cache $0.14 | $4.18 | $0.665 agentic 92/8 | 1M | Mistral's largest model, multimodal agents |
| Mistral Large 3 | $0.50 cache $0.05 | $1.50 | $0.241 −2.8× | documented elsewhere — not on pricing page | The previous Large, open weights out now |
| Mistral Medium 3.5 | $1.50 cache $0.15 | $7.50 | $0.962 +45% | documented elsewhere — not on pricing page | Mistral's agentic and coding model |
| GPT-6.1 Sol | $2.00 cache $0.10 | $10.00 | $1.29 +94% | 1.05M | OpenAI balanced model for agents |
| Claude Sonnet 5.5 | $2.00 cache $0.10 | $10.00 | $1.29 +94% | 1M | Claude's speed-and-intelligence default |
| Claude Haiku 5.5 | $0.10 cache $0.01 | $0.50 | $0.0682 −9.7× | 1M | Anthropic small model for volume |
By comparison
Frequently asked.
What Mistral Large 4 costs now and after the launch sale, what the preview label means, and how it sits next to Large 3.
Q · 01 How much does Mistral Large 4 cost? +
The list price on Mistral's API pricing page and model card is
$1.36/M input, $0.14/M cached input and $4.18/M output. During the launch sale Mistral charges half: $0.68, $0.07 and $2.09 per million tokens. The page marks the sale figure with a 'Sale price' badge whose tooltip calls it a temporary sale price and the struck-through amount the original price, so this page and its calculator use the list figure.Q · 02 When does the launch sale end? +
Mistral's changelog entry for October 6, 2026 says 'Launch pricing: 50% off for 2 weeks', which points to around October 19-20, 2026. The pricing page itself prints no end date. We will recheck the card when the two weeks are up rather than assume the list price has returned.
Q · 03 Is Mistral Large 4 open-weight? +
Not yet. Mistral describes it as an open-weight model, but the changelog says 'Open weights are coming soon' and the launch post says the weights will be released by the end of the month. Until then Mistral is red-teaming it with security partners and state authorities, and the only way to run it is Mistral's API. Mistral Large 3 is the Large whose weights you can download today.
Q · 04 What does 'public preview' mean for production use? +
The model is on the paid API with a published price, but Mistral labels it a preview and says it 'continues to improve rapidly as we refine it', so behaviour can change under the same model ID,
mistral-large-4. Mistral publishes no retirement or GA date for the preview yet.Q · 05 How does it compare with Mistral Large 3 on price? +
Large 3 lists at
$0.50/M input and $1.50/M output, so Large 4's list card is about 2.7x on input and 2.8x on output, and even the sale price is above Large 3. Under this site's cache-heavy blend the figures are $0.665/M here against $0.241/M. Large 3 has a 256K window and downloadable weights; Large 4 has 1M and a larger model behind it.Q · 06 Is there a Batch or Priority rate? +
Yes. The Batch tab on Mistral's pricing page lists Large 4 at
$0.68/M input and $2.09/M output at list, half the standard card, and the Priority tab at $2.38 and $7.315 - 1.75x standard. The launch sale halves both tiers as well.Q · 07 What context window and inputs does it take? +
The model card lists a 1M-token context window and text and image input. The architecture is a Mixture-of-Experts with 52B active parameters, about 1T in total, and a 1.6B-parameter vision encoder. Features listed on the card: structured outputs, function calling, document Q&A, prefix, batching, agents and conversations, and built-in tools.
Q · 08 How does it score? +
Mistral's launch post reports 61.7% on DeepSWE v1.1, 59.4% on SWE-Atlas-QnA, 28.3% on Terminal-Bench 4 and 59.9% on AutomationBench, and says it ranks among the top five models on the Artificial Analysis Cyber Index. These are Mistral's own figures for a preview that it says is still improving, not independent results.