Mistral Large 3 675B Instruct vs Mistral Medium 3.5

On shared benchmark coverage, MMLU PRO has Mistral Large 3 675B Instruct at 85.5 and Mistral Medium 3.5 at 79.8, with Mistral Large 3 675B Instruct ahead by 5.7 points; Google-Proof Q&A has Mistral Large 3 675B Instruct at 43.9 and Mistral Medium 3.5 at 71.3, with Mistral Medium 3.5 ahead by 27.4 points; τ-bench has Mistral Large 3 675B Instruct at 70.2 and Mistral Medium 3.5 at 91.4, with Mistral Medium 3.5 ahead by 21.2 points. The largest visible gap is 27.4 points on Google-Proof Q&A, which matters most when that benchmark mirrors your workload. Treat isolated benchmark wins as directional, because provider routing, prompt style, and tool access can move real application results.

The capability footprint differs most on reasoning mode: Mistral Medium 3.5, function calling: Mistral Medium 3.5, and tool use: Mistral Medium 3.5. Both models share vision, multimodal input, and structured outputs, so the practical split is not just feature count. Use those differences to decide whether the page is about raw model quality, agentic coding support, multimodal ingestion, or predictable structured API behavior.

For cost, Mistral Large 3 675B Instruct lists $0.50/1M input and $1.50/1M output tokens on the cheapest tracked provider, while Mistral Medium 3.5 lists $1.50/1M input and $7.50/1M output tokens on the cheapest tracked provider. A 70/30 input-output blend puts Mistral Large 3 675B Instruct lower by about $2.50 per million blended tokens. Availability is 5 providers versus 3, so concentration risk also matters.

Choose Mistral Large 3 675B Instruct when vision-heavy evaluation, lower input-token cost, and broader provider choice are central to the workload. Choose Mistral Medium 3.5 when reasoning depth and larger context windows are more important. For production, rerun your own prompts through the exact provider, region, and tool stack you plan to ship.