█

AI/TLDR

Buy me a coffee
Buy me a coffee

Mistral Large 4

Mistral's 1T-parameter multimodal MoE flagship with 52B active parameters and a 1M-token context

IN SHORT

What is Mistral Large 4?

Mistral Large 4 is Mistral AI's flagship model, launched as a public preview on October 6, 2026. It is a natively multimodal mixture-of-experts (MoE) model with 1.05 trillion total parameters, 52 billion active parameters and a 1.6B vision encoder, and Mistral's documentation lists a 1M-token context window. It has a 1M-token context window and costs $0.70 per million input tokens and $2.10 per million output tokens.

Older version
Mistral Large 3 · 2 December 2025
Where to use it
Mistral AI (Mistral Studio)
Updated Checked against 3 sourcesMarkdown version ↗
Cite this pageAI/TLDR, "Mistral Large 4: specs, pricing and benchmarks", https://ai-tldr.dev/models/mistral-large-4/ (updated 6 October 2026)
// FACTSas of 6 October 2026
Context window1M[1]
Input price$0.70 / 1M tokenscached $0.07[2]
Output price$2.10 / 1M tokens[2]
LicenceMistral Ecosystem Open-Weight License (modified MIT) · open weights[1]
Released6 October 2026[3]
Parameters1.05T total · 52B active[1]
  1. [1] huggingface.co Weights on Hugging Face
  2. [2] docs.mistral.ai Mistral Large 4 docs
  3. [3] mistral.ai Introducing Mistral Large 4 (announcement)
Mistral LargeOpen weightsGenerally available (API) + open weights
Released
6 Oct 2026
Context
1M
Parameters
1.05T total · 52B active
Input
$0.70 / 1M tokens
License
Mistral Ecosystem Open-Weight License (modified MIT)
Coverage
2 stories

Overview

Mistral Large 4 is Mistral AI's flagship model, launched as a public preview on October 6, 2026. It is a natively multimodal mixture-of-experts (MoE) model with 1.05 trillion total parameters, 52 billion active parameters and a 1.6B vision encoder, and Mistral's documentation lists a 1M-token context window. Mistral calls it "ML4" unofficially, and "le Chonk" very officially.

Mistral trained the model on 3,800 NVIDIA Grace Blackwell GPUs in its own European datacenters. The launch post focuses on coding, agentic workflows and multimodal understanding, with extra attention on cybersecurity, finance and legal work, and says the model supports more than 160 languages, including all official EU languages. It reports 61.7% on DeepSWE v1.1, 59.4% on SWE-Atlas-QnA, 28.3% on Terminal-Bench 4, 59.9% on AutomationBench and 93% of the 40 Cybench challenges.

The API is available on Mistral Studio, with structured outputs, function calling, document Q&A and batching. On October 29, 2026 Mistral released the weights on Hugging Face (FP8 and NVFP4 checkpoints, served with vLLM or SGLang) under the Mistral Ecosystem Open-Weight License, a modified MIT license that is free to use, fine-tune and commercialize except for Model-as-a-Service businesses, which need a separate license.

The open-weight release is an improved model: three more weeks of reinforcement learning lifted Terminal-Bench 4.0 from 28% to 41% with 30% fewer output tokens, and it scores 48% on the Artificial Analysis Coding Agent Index, effectively tied with GLM-5.3. Reasoning effort is set per request to low, medium or high (the default).

Mistral's documentation lists Mistral Large 4 as generally available from November 5, 2026, served under the API names mistral-large-4, mistral-large-4-0 and mistral-large-latest at unchanged pricing.

Released2026-10-06
LicenseMistral Ecosystem Open-Weight License (modified MIT)
WeightsOpen weights
Parameters1.05T total · 52B active
Context1M
ArchitectureMixture-of-Experts
ModalitiesText, Vision
StatusGenerally available (API) + open weights

Benchmarks

Mistral's published bar chart of Artificial Analysis DeepSWE 1.1 scores: Mistral Large 4 Preview 62, Beam 44, Qwen3.8 Max 51, DeepSeek-V4-Pro-0813 57, GLM-5.3 61 and Kimi K3 68.
Artificial Analysis DeepSWE 1.1 results for the Mistral Large 4 preview, as published by Mistral (October 2026). — Mistral AI
Mistral's published bar chart of Artificial Analysis Terminal-Bench 4 scores: Mistral Large 4 Preview 28, DeepSeek V4 Pro 0813 10, Qwen3.8 Max 17, Kimi K3 21 and GLM-5.3 40.
Artificial Analysis Terminal-Bench 4 results for the Mistral Large 4 preview, as published by Mistral (October 2026). — Mistral AI

Surge AI human evaluation of coding quality (score out of 5), as published by Mistral

BenchmarkMistral Large 4 PreviewClaude Opus 5GLM-5.3Kimi K3GLM-5.2
Surge AI coding human eval3.74 /54.22 /53.6 /53.59 /53.4 /5

Comparison source ↗

Pricing

Input$0.70 / 1M tokens
Cached input$0.07 / 1M tokens
Output$2.10 / 1M tokens

Pricing source ↗

Strengths

  • Frontier-scale MoE: 1.05T total parameters with only 52B active per token
  • 1M-token context window on the API
  • Native vision through a 1.6B vision encoder
  • Strong security results: 93% of Cybench challenges and 82% on the AA Cyber Index vulnerability-reproduction test
  • Open weights under a modified MIT license, released October 29, 2026

Best for

  • Coding agents and long terminal workflows
  • Agentic business automation
  • Security work such as vulnerability reproduction
  • Finance and legal document analysis across many languages

How to access

ProviderModel ID
Mistral AI (Mistral Studio) ↗mistral-large-4

Mistral Large — every version

The full lineage of the Mistral Large line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.

VersionReleasedContextLicense
Mistral Large 4current2026-10-061M—
Mistral Large 32025-12-02256KApache-2.0
Mistral Large 2.1 (24.11)2024-11-18—Open weights
Mistral Large 2 (24.07)2024-07-24—Open weights
Mistral Large (24.02)2024-02-26—Proprietary

FAQ

Is Mistral Large 4 open-weight?

Yes. Mistral Large 4 launched as an API-only public preview on October 6, 2026, and Mistral released its weights on Hugging Face on October 29, 2026 under the Mistral Ecosystem Open-Weight License. That modified MIT license allows free commercial use and fine-tuning, but Model-as-a-Service providers need a separate commercial license.

How much does Mistral Large 4 cost?

Mistral Large 4 costs $0.70 per million input tokens, $0.07 per million cached input tokens and $2.10 per million output tokens on the Mistral API, per Mistral's docs page. The weights themselves are free to download from Hugging Face for self-hosting.

How big is Mistral Large 4?

Mistral Large 4 is a mixture-of-experts model with 1.05 trillion total parameters and 52 billion active parameters per token, plus a 1.6B vision encoder. It is larger than Mistral Large 3 (675B total, 41B active) and its context window grows from 256K to 1M tokens.