Models · Compare
Mistral Large 3 vs Claude Opus 4.5
Rows highlighted in warm gray are where the models differ. Numbers carry their as-of date and primary source.
Specs
| Field | A: Mistral Large 3 | B: Claude Opus 4.5 |
|---|---|---|
| Released | 2025-12-02 | 2025-11-24 |
| Developer | Mistral AI | Anthropic |
| Openness | Open weights | Proprietary |
| License | Apache-2.0 | Proprietary |
| OSI-approved | yes | no |
| Data released | no | no |
| Training code | no | no |
| Architecture | moe | unknown |
| Total params | 675B | — |
| Active params | 41B | — |
| Experts | — | — |
| Context window | 256K | — |
| Attention | unknown | unknown |
| Position enc. | unknown | unknown |
| Pretraining tokens | — | — |
| Post-training | sft, rlhf | rlhf, constitutional |
| Training hardware | H200 | — |
| $/M input | $0.50 | $5.00 |
| $/M output | $1.50 | $25.00 |
| Output tok/sec | 54.2 | 51.8 |
Benchmarks
Missing scores render as not reported; never inferred. Bold highlights the leader per benchmark.
General reasoning
| MMLU-Pro | 80.7 2026-05-21 | 88.9 2026-05-21 |
| GPQA-Diamond | 68.0 2026-05-21 | 81.0 2026-05-21 |
Code
| LiveCodeBench | 46.5 2026-05-21 | 73.8 2026-05-21 |
Math
| AIME 2025 | 38.0 2026-05-21 | 62.7 2026-05-21 |
Context · A
Mistral's first MoE since the Mixtral series, released December 2 2025 as base and instruct under Apache 2.0. 675B total / 41B active, trained from scratch on 3000 NVIDIA H200 GPUs. Multimodal with a 256K context, distributed day one on Amazon Bedrock, Azure Foundry, and Hugging Face.
Context · B
Anthropic's first Opus to cross 80 percent on SWE-Bench Verified per the lab's own numbers, released November 24 2025 at a two-thirds price cut versus Opus 4.1 ($5 / $25 per Mtok). Added an effort parameter for adjustable reasoning intensity, with medium-effort runs matching Sonnet 4.5 using 76 percent fewer output tokens.