Mistral Large 4
Mistral's 1-trillion-parameter open-weight flagship, nicknamed 'le Chonk'
Context window
1M
Input / 1M tokens
$1.36
Output / 1M tokens
$4.18
Provider
Mistral AI
Cached input $0.14 per 1M. Mistral docs also list a 50% lower rate ($0.68/$2.09) that Artificial Analysis describes as a launch discount. OpenRouter serves a 512K context window. · Data verified 2026-10-07
Mistral Large 4 (October 2026) is a natively multimodal Mixture-of-Experts model with about 1 trillion total parameters and 49 billion active per token (Mistral's docs list 1.05T total and 52B active). Mistral calls it its largest and most capable model and pitches it on European AI sovereignty, with private-cloud and on-premises deployment. It opened as a public preview on October 6; open weights are promised for late October, with the license not yet announced. Independent testing places it as the strongest model built outside the US and China, though still well behind the closed frontier.
Capability index
Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.
How to access it
Public preview on the Mistral API (model ID mistral-large-4-0, alias mistral-large-4), served from EU infrastructure. Open weights are expected by the end of October 2026; the license has not been announced.
Strengths
- ✓Open weights promised: self-host a 1T-parameter model on your own infrastructure
- ✓Low API price for its size ($1.36/$4.18 per 1M tokens)
- ✓Strong cyber results: 93% on Cybench, 82% on a vulnerability reproduce-and-patch test (Mistral-reported)
- ✓Trained on 160+ languages including all EU official languages
- ✓EU-hosted inference for data sovereignty requirements
Best for developers who...
When to choose it (and when not to)
Reach for Mistral Large 4 when...
- →You need an open-weight frontier-class model you can eventually self-host
- →EU data residency or sovereignty is a hard requirement
- →You want strong multilingual European-language coverage at a low token price
Look elsewhere if...
- ✕You need the strongest available coding or reasoning model
- ✕You need a finalized license today (it has not been published yet)
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| DeepSWE v1.1 | 61.7% | Mistral-reported |
| Terminal-Bench 4 | 28.3% | Mistral-reported |
| AutomationBench | 59.9% | Mistral-reported |
| Cybench | 93% | Mistral-reported |
| Artificial Analysis Intelligence Index | 38 | Third-party; highest outside the US and China, vs 58 for Claude Opus 5.5 |
Source: Mistral AI announcement
Compare Mistral Large 4
Compare Mistral Large 4 with any other model
Build a comparison →All model comparisons →