Mistral Large 4 is a 1 trillion-parameter natively multimodal model with 49 billion active parameters, built as a hybrid instruct-and-reasoning Mixture of Experts. Mistral describes it as its largest and most capable model to date, and the weights are open.
| Developer | Mistral AI |
|---|---|
| Model ID | mistralai/mistral-large-4-0 |
| Modalities | Image, Text → Text |
| Context window | 524,288 tokens |
| Max output | 262,144 tokens |
| Parameters | 1 trillion total, 49 billion active |
| Architecture | Hybrid instruct-and-reasoning Mixture of Experts |
| Released | 6 October 2026 |
Sparse at scale
1 trillion parameters in total but 49 billion active per token, so capacity does not translate one-for-one into cost per request.
A 512K window
524,288 tokens in and up to 262,144 out, which fits an entire service — source, tests and a ticket thread — in one prompt.
Instruct and reasoning in one
A hybrid design rather than separate chat and reasoning checkpoints.
Native multimodality
Image input is built in rather than bolted on.
Published by Mistral AI with the Mistral Large 4 announcement, retrieved 6 October 2026. These are the vendor's own figures and have not been independently reproduced.
Long-running agent work
The DeepSWE and AutomationBench figures are both about finishing multi-step tasks rather than answering one question.
Whole-repository reasoning
A 512K window keeps the context in the prompt instead of in a retrieval layer.
Sources: https://mistral.ai/news/mistral-large-4 · https://docs.mistral.ai/models/mistral-large-4-0
Measured on the AI/ML API gateway across live traffic. Figures are medians unless stated. Last updated 2026-10-12.
import requests r = requests.post( "https://api.aimlapi.com/v1/chat/completions", headers={"Authorization": "Bearer " + AIMLAPI_KEY}, json={ "model": "mistralai/mistral-large-4-0", "messages": [ { "role": "user", "content": "Hello!" } ] }, ) print(r.json())
const r = await fetch("https://api.aimlapi.com/v1/chat/completions", { method: "POST", headers: { Authorization: "Bearer " + process.env.AIMLAPI_KEY, "Content-Type": "application/json", }, body: JSON.stringify({ "model": "mistralai/mistral-large-4-0", "messages": [ { "role": "user", "content": "Hello!" } ] }), }); console.log(await r.json());
curl -X POST https://api.aimlapi.com/v1/chat/completions \ -H "Authorization: Bearer $AIMLAPI_KEY" \ -H "Content-Type: application/json" \ -d '{"model":"mistralai/mistral-large-4-0","messages":[{"role":"user","content":"Hello!"}]}'
OpenAI-compatible — swap the base URL and it works with your existing SDK.
| Type | Price |
|---|---|
| Input | |
| Output | |
| Cached input |
Billed per token, separately for input and output.
| Benchmark | Score | What it measures | Source | Retrieved |
|---|---|---|---|---|
| Intelligence | 38.4 | Composite score across standardised reasoning, knowledge and problem-solving evaluations, measured independently by Artificial Analysis | Source | October 6, 2026 |
| DeepSWE | 61.7% | Agentic software engineering: resolving real repository tasks end to end | Source | October 9, 2026 |
| SWE-Atlas Codebase QnA | 59.4% | Answering questions about an unfamiliar codebase | Source | October 9, 2026 |
| Terminal-Bench 4 | 28.3% | Autonomous shell and terminal task completion, fourth release | Source | October 9, 2026 |
| AutomationBench | 59.9% | Completion of routine automation workflows | Source | October 9, 2026 |
Figures published by Mistral AI with the Mistral Large 4 announcement (mistral.ai/news/mistral-large-4), retrieved 2026-10-06: DeepSWE v1.1 61.7%, Cybench 93%, AutomationBench 59.9%, Dense 200 visual grounding 42%, B3 Attack Resistance 93.3%, KORA 1.691.
| Model | Input | Output | Context | Best for |
|---|---|---|---|---|
Mistral Large 4 This page | Chat + assistants | |||
| Chat + assistants | ||||
| Reasoning + agents | ||||
| Long-form output and agentic software engineering |
Mistral Large 4 is a 1 trillion-parameter natively multimodal model with 49 billion active parameters, built as a hybrid instruct-and-reasoning Mixture of Experts. Mistral describes it as its largest and most capable model to date.
$1.768 per 1M input tokens and $5.434 per 1M output tokens through AI/ML API.
Yes. Mistral describes it as natively multimodal, taking both text and image input and returning text.
With the announcement Mistral reported DeepSWE v1.1 61.7%, Cybench 93% on vulnerability reproduction and patching, AutomationBench 59.9%, Dense 200 visual grounding 42%, B3 Attack Resistance 93.3%, and a KORA score of 1.691. These are the vendor's own figures, not independently reproduced.
Mistral describes the model as open-weight and said at announcement that the weights would drop at the end of the month.
