Mistral’s lineup is not a single interchangeable model. Choose by workload and deployment status, then verify the exact endpoint and price. Large 4 is the current flagship public preview, not a downloadable local model as of October 7. Older model guides remain useful for their specific generations; they should not be mistaken for the current flagship.

Current flagship: Mistral Large 4

The October 6 launch introduced a native multimodal mixture-of-experts model for coding, agents and document reasoning. The API is usable now through Mistral Studio with model ID mistral-large-4; weights are announced for the end of October.

The live model card lists a 1M-token context and introductory input/cached-input/output prices of $0.68/$0.07/$2.09 per million tokens. Displayed list rates are $1.36/$0.14/$4.18. Mistral’s changelog describes a two-week launch discount. Those rates are not a permanent pricing promise.

Active parameters remain inconsistent across official sources: 49B in the launch announcement versus 52B in both live model-card routes checked October 7. We do not silently choose a number or infer local hardware needs. See the Large 4 pricing, specs and availability reference for the full distinction.

Pick by workload, not by the largest number

Requirement	Starting point	What to verify
New flagship vision, reasoning or tool workflows	Large 4 preview	Preview behavior, introductory versus list pricing, task reliability
Existing Medium integration	Medium 3.5 guide	Exact model ID, current provider availability and rates
Existing API application	Mistral API guide	Endpoint, account limits, lifecycle and migration support
Coding-agent evaluation	Agent model guide	Completed-task cost and tool-use failure modes
Offline or self-hosted coding	Open-weight coding guide	Actual weights, license and supported runtime

A strong vendor benchmark is a reason to evaluate, not a reason to silently replace an established production model. Keep prompts, tools and scoring comparable when testing the flagship preview.

Earlier models and historical references

Our Large 2 guide, Devstral 2 guide and Codestral guide describe specific earlier products and use cases. Do not use their old prices, local-memory estimates or benchmarks as Large 4 specifications. API retirement and downloadable-weight availability are separate lifecycle questions.

For current provider economics, use the Get API Pulse Mistral pricing reference and cost calculator. Model versions remain separate there, so an existing Large 3 integration is not relabeled Large 4.

Sources and availability boundary

Checked October 7, 2026: Mistral launch announcement, current model card, versioned card and changelog.

Until official Large 4 weights, licensing and deployment instructions exist, this directory makes no self-hosting, GPU-memory or local-install recommendation for it.