Mistral Large 4
A European flagship with 1T open weights and a $0.7 input price, built for coding and cyber defence
Benchmark scores
How it is measured →Ranked 19 · bars are scaled to the field maximum for that benchmark.
| Benchmarks | Score | Vs. best in field | Gap |
|---|---|---|---|
| AA Intelligence Index v5 Intelligence index | 26.0 | 43% | −34.0 pts |
| AA Intelligence Index v4.3.2 | — | — | not indexed |
| GPQA Diamond Graduate-level science reasoning | 87.6 | 91% | −8.8 pts |
| Terminal-Bench 4.0 | — | — | not indexed |
| LiveBench Design | — | — | not indexed |
| Vals Index Vals composite index | 56.53 | 77% | −16.6 pts |
Assessment
Full weights released 2026-10-29 in FP8 and NVFP4 (Hugging Face: Mistral-Large-4-1T-A52B). A 1T-total / 52B-active MoE with 1M context, image input and 160+ languages; Mistral recommends eight B300 GPUs to serve it. AA intelligence index 26, fifth among open weights; vendor-reported 19.7% on FrontierCode 2.0 is the best open score; Terminal-Bench 4.0 is vendor-reported at 41% but Vals measured 26.3%, and AA has not listed it — so the table leaves it blank. Priced $0.70/$2.10 with $0.07 cache. The licence is modified MIT: offering it as MaaS commercially needs a separate agreement.
At the same price
The four models closest on blended price, side by side.
| Model | v5 | Blended | tok/s | Score gap |
|---|---|---|---|---|
| MiMo-V2.6 Pro Open weightsXiaomi | 30.0 | $0.61 | 81 | +4.0 |
| DeepSeek V4.1 Flash Open weightsDeepSeek | 33.0 | $0.53 | 148 | +7.0 |
| MiniMax M3.1 FlashMiniMax | 29.0 | $0.26 | 118 | +3.0 |
| Claude Haiku 5.5Anthropic | 36.0 | $0.20 | 235 | +10.0 |
Compare side by side
Line it up against other models on price, speed and every benchmark.
The comparison tool lines up to 4 models side by side with switchable normalisation — built for purchasing decisions.
Open the comparison tool → Back to leaderboard