On 2 December 2025 Mistral released the Mistral 3 family, including Mistral Large 3 and smaller Ministral variants. The company’s announcement explicitly states Apache 2.0 licensing for the release. That is more deployment-flexible than a merely accessible hosted API, but it also transfers operational responsibilities to the organization running the model.
What you actually receive
Mistral Large 3 is described as a sparse mixture-of-experts model with 675B total and 41B active parameters. Smaller dense options target tighter hardware budgets. Before planning inference capacity, collect exact weight filenames, precision, intended processor configurations and context recommendations from the published card. A reported active-parameter count should never be used as the sole GPU-memory estimate.
Permissive does not mean turnkey
Apache 2.0 supplies a familiar software licensing framework for the announced weights, subject to the actual shipped files. It does not certify the training data, remove all privacy risk or provide production moderation, logging and tenancy isolation. Self-hosting can improve control of where prompts are processed, but the owner must configure storage, access controls, monitoring and patching.
Tool calling and JSON
Use the model’s specified chat template and tool-call parser. A checkpoint’s general instruction-following strength is not sufficient evidence that it supports the exact structured schema a framework sends. Create integration tests for missing fields, malformed arguments, retries, streaming and cancellation.
Comparative value
If cloud GPU overhead is significant, a hosted alternative may still be operationally cheaper; that requires current rates from AI 10. For teams that require local weight inspection, Mistral 3 is a verified historical choice whereas Mistral Large 4’s 6 October 2026 announcement was initially an API preview awaiting weights.
What to verify before adopting this model or method
Use the exact release ID and API endpoint; confirm your geographic processing location, entitlement tier, model lifecycle, feature flags, context and output limits from the provider console. Run a reproducible evaluation on your own data and retain human oversight for consequential actions. Where a specification, price, commercial license or entitlement is not verified for your deployment, this article deliberately leaves it unasserted instead of filling a plausible number.
Sources & Further Reading
- mistral.ai — documentation/reference (supplier-provided reference; not independently checked in this pass)
- mistral.ai — documentation/reference (supplier-provided reference; not independently checked in this pass)
Source documentation establishes vendor-published facts, not independently measured model quality, tenant access, contractual rights or current prices. Use a dated pricing ledger and an account-specific entitlement check for deployment decisions.