MISTRAL LARGE 3

Mistral Large 3: Europe's 675B flagship

675B parameters with vision. 256K context. Top-tier multilingual reasoning. $0.80/$2.40 per million tokens — a fraction of GPT-5.5 and Opus 4.7.

Try Mistral Large 3 on Council See it in a council

Mistral Large 3 is Mistral AI's flagship and the strongest open-weights-adjacent frontier model coming out of Europe in 2026. With 675B parameters, native vision input, and a 256K context window, it competes with GPT-5.5 and Claude Opus 4.7 on reasoning and multilingual quality while costing a fraction of either — $0.80 per million input tokens and $2.40 per million output. It's particularly strong on French, German, Spanish, Italian, and Arabic, where the closed-weight US labs sometimes lose nuance. The case for: data-residency requirements (Mistral is EU-headquartered), multilingual product work, and any team that wants frontier quality without paying frontier prices. The case against: agent loops where GPT-5.5's tool-calling stability still wins, and SWE-bench coding work where Claude Opus 4.7 leads. Native web search is not enabled in Council AI for Mistral models — Mistral's web search requires the Agents API, which the platform does not yet integrate.

Specs

ProviderMistral AI
Parameters675B
Context window256,000 tokens
MultimodalText + image input; text output
Web searchNot enabled in Council (requires Agents API)
Price (in / out per 1M)$0.80 / $2.40
StrengthsMultilingual, reasoning, vision, EU residency

Best at

Where it loses

How to use it in a council

Pair Mistral Large 3 with GPT-5.5 (for tool calling) and Claude Sonnet 4.6 (for coding) in a 3-model council. You get European multilingual coverage, US tool-calling reliability, and Anthropic's code quality — all synthesized into a single response. The cheap Mistral input price keeps the council's overall cost low even when one of the other two is more expensive.

Frequently asked questions

Is Mistral Large 3 open-weights?

Large 3 itself is a hosted API model, but Mistral has historically released open-weight versions of prior flagships. Council AI uses the hosted API for production reliability.

How much does Mistral Large 3 cost?

$0.80 per million input tokens and $2.40 per million output. Roughly 6x cheaper input than GPT-5.5 and 12x cheaper output. Council AI bundles it inside monthly plan budgets.

Why doesn't Mistral have web search in Council?

Mistral's web search lives in their Agents API, which is a separate orchestration layer. Council AI doesn't integrate it yet. If you need fresh data, route the same query to Gemini 3 Pro or GPT-5.5 in the same council.

Is the 256K context real or interpolated?

Native 256K. Mistral Large 3 holds coherence across the full window better than the prior Large 2 generation.

When should I pick Mistral Large 3 over GPT-5.5?

Multilingual products (especially French/German/Spanish), price-sensitive frontier work, EU data residency requirements, or when you specifically want a non-US model in your council.

Can Mistral Large 3 read images?

Yes — native vision input. Upload screenshots, diagrams, or charts and ask about them. Council AI handles the routing.