Mistral Large 4 is Mistral's open-weight AI model, released 29/10/2026. Mistral Large 4 is a natively multimodal mixture-of-experts model with about 1.05T total and 52B active parameters, released under the Mistral Ecosystem Open-Weight License (a modified MIT licence that needs a separate commercial licence for model-as-a-service use). The launch post is strongest on cybersecurity and agent-security results and prints few comparison numbers, with several charts (ChartQA Pro, SciCode-Verified, GDP.pdf, finance and legal evaluations) showing no readable scores, so those rows are left out. The docs page lists 5 November 2026 as its release date for the model (public preview began 6 October 2026). The model card and launch post disagree on Terminal-Bench 4 (41 against 28.3).
Visit Mistral Large 4AI-Powered
Leverages advanced AI technology to deliver cutting-edge capabilities and results.
Fast & Efficient
Optimized performance ensures quick results without compromising on quality.
Purpose-Built
Specifically designed for llms tasks and workflows.
How does Mistral Large 4 score on independent benchmarks?
Mistral Large 4 ranks #24 of 28 models on the Artificial Analysis Intelligence Index (v5.0.1) with a score of 26.1, as measured on 22 November 2026.
Where does Mistral Large 4 rank against other models?
How good is Mistral Large 4 at agentic coding?
How does Mistral Large 4 do on individual tests?
| Evaluation (run by Artificial Analysis) | Mistral Large 4 | Best of any model |
|---|---|---|
| Terminal-Bench 4.0 | 24.7% | 67.2% |
| FrontierCode 2.0 | 15% | 48.8% |
| Humanity's Last Exam | n/a | 74.3% |
| GDPval-AA (Elo) | n/a | 1866 |
How has Mistral Large 4's score changed over time?
We have tracked Mistral Large 4 on Intelligence Index v5.0.1 since 22/11/2026. A point is added each day the score, coding index or price changes, so this chart fills in over time.
What does the Mistral Large 4 system card report?
Mistral Large 4 is a natively multimodal mixture-of-experts model with about 1.05T total and 52B active parameters, released under the Mistral Ecosystem Open-Weight License (a modified MIT licence that needs a separate commercial licence for model-as-a-service use). The launch post is strongest on cybersecurity and agent-security results and prints few comparison numbers, with several charts (ChartQA Pro, SciCode-Verified, GDP.pdf, finance and legal evaluations) showing no readable scores, so those rows are left out. The docs page lists 5 November 2026 as its release date for the model (public preview began 6 October 2026). The model card and launch post disagree on Terminal-Bench 4 (41 against 28.3).
These are Mistral AI's own published results (self-reported, with the vendor's harness and settings), so they are not directly comparable with other labs' numbers. Sources: Mistral Large 4 announcement (public preview) (06/10/2026), mistralai/Mistral-Large-4.0-1T05-A52B on Hugging Face, Mistral docs: mistral-large-4-0. Last checked 22/11/2026.
Coding
| Benchmark | Mistral Large 4 | Claude Opus 5 | GLM-5.3 | Kimi K3 |
|---|---|---|---|---|
| DeepSWE v1.1Launch post chart | 61.7% | n/a | n/a | n/a |
| SWE-Atlas-QnALaunch post chart | 59.4% | n/a | n/a | n/a |
| Terminal-Bench 4Launch post chart (the Hugging Face card lists 41 on the same benchmark name) | 28.3% | n/a | n/a | n/a |
| Terminal-Bench 4 (Hugging Face card)Value listed on the model card metadata, which differs from the launch post chart | 41% | n/a | n/a | n/a |
| Coding Agent Index (Artificial Analysis)Launch post: ahead of DeepSeek V4 Pro 0813 and Qwen3.8 Max (scores not printed) | 49.8% | n/a | n/a | n/a |
| Coding Agent Index (Hugging Face card)Card also states 47% fewer tokens per task on average | 48% | n/a | n/a | n/a |
| CWE-Bench resolve rateValue listed on the model card | 46.3% | n/a | n/a | n/a |
| Terminal-Bench Science v0.1Value listed on the model card | 4.3% | n/a | n/a | n/a |
| Surge AI blind human coding evaluationPreview model; second of five models tested | 3.74 points (1 to 5) | 4.22 | 3.6 | 3.59 |
Agents & tool use
| Benchmark | Mistral Large 4 | Claude Opus 5 | GLM-5.3 | Kimi K3 |
|---|---|---|---|---|
| AutomationBench (657 business workflows)Launch post: ahead of Kimi K3, MiMo-V2.6-Pro and DeepSeek V4 Pro (scores not printed) | 59.9% | n/a | n/a | n/a |
| AA-BriefcaseLaunch post: ahead of DeepSeek V4 Pro (score not printed) | 1393 Elo | n/a | n/a | n/a |
| CybenchShare of 40 challenges solved | 93% | n/a | n/a | n/a |
| Artificial Analysis Cyber Index: reproduce-and-patch real vulnerabilitiesHighest of any model on the test per the launch post | 82% | n/a | n/a | n/a |
Multimodal
| Benchmark | Mistral Large 4 | Claude Opus 5 | GLM-5.3 | Kimi K3 |
|---|---|---|---|---|
| Dense 200 (visual grounding) | 42% | n/a | n/a | n/a |
Safety & honesty
| Benchmark | Mistral Large 4 | Claude Opus 5 | GLM-5.3 | Kimi K3 |
|---|---|---|---|---|
| B3 AI Security Benchmark (Lakera)Lakera public weak-agent dataset | 93.3 % of attacks resisted | n/a | n/a | n/a |
| KORABenchHighest among open-source models measured, per the launch post | 1.691 points (maximum 2, Exemplary) | n/a | n/a | n/a |
Other
| Benchmark | Mistral Large 4 | Claude Opus 5 | GLM-5.3 | Kimi K3 |
|---|---|---|---|---|
| Artificial Analysis Intelligence Index v5Value listed on the model card | 26 points | n/a | n/a | n/a |
What did safety testing find for Mistral Large 4?
Safety level: No framework rating published; Mistral reports security and alignment benchmarks only
- Resists 93.3% of attacks on Lakera's B3 AI Security Benchmark.
- Scores 1.691 out of 2 on KORABench, the highest among open-source models measured.
- Cyber refusal rate on JailbreakBench, StrongREJECT and AgentHarm is described as higher than all open-source models, with no figure given.
Independent figures: Artificial Analysis, refreshed daily on our AI model benchmarks page. Full analysis: our Mistral Large 4 deep dive.
Mistral Model Timeline
1M tokens context
262k tokens context
Specifications
Related Tools
Claude Opus 5.5
Claude Opus 5.5, launched 22 September 2026, is Anthropic's current Opus-line flagship, replacing Opus 5. It performs at roughly the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5, at $4 input / $20 output per million tokens (both 20% down, cache reads cut 60% to $0.20). It leads on Terminal-Bench 4.0, GDPval-AA v2.1 and Humanity's Last Exam, and is the first Opus model to ship with Fable-5.1-class cybersecurity, biology and distillation safeguards.
Claude Sonnet 5.5
Claude Sonnet 5.5, released 28 September 2026, is the second model in Anthropic's Claude 5.5 family and the faster, lower-cost complement to Claude Opus 5.5. It keeps Sonnet 5's $2 input / $10 output per million tokens ($0.20 cache reads) while generating output over 30% faster, and scores 70.6% on Terminal-Bench 4.0, above Opus 5.5's 66.4%. It has a 1M-token context window, 128K max output, adaptive thinking, and is the first Sonnet with cyber and anti-distillation classifiers.
GPT-6.1 Sol
GPT-6.1 Sol, launched at OpenAI DevDay on 29 September 2026, is OpenAI's mid-tier workhorse model. OpenAI says it delivers nearly the same intelligence as GPT-6 Astra at one-fifth of Astra's token price: $2 input / $10 output per million tokens with cached input cut to $0.10. It scores 75.2% on DeepSWE v1.1 (high effort) and 71.4% on OSWorld 2.0, has a 1.05M-token context window and 128K max output, and is available in the API plus ChatGPT Work and Codex.
