Leanstral 1.5

Pricing — 1 offering(s)

Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.

Capability profile

Operator guidance

Not a general-purpose chat, reasoning, or coding model — route to this only for Lean 4 proof engineering, automated theorem proving, or formal verification tasks specifically. For general coding, use Codestral or another general-purpose Mistral model; for general reasoning or chat, use Mistral Large or Small. This model's entire value proposition is the narrow domain it's built for.

Use cases

• Automated theorem proving and proof engineering in Lean 4
• Autoformalization — translating informal mathematical statements into formal Lean 4 proofs
• Formal verification of code properties, including bug-finding in existing repositories via rigorous proof rather than pattern-matching

Limitations

• Not a general-purpose coding or chat model — capability data here deliberately does not cover general coding/reasoning/chat quality, since the model isn't built or evaluated for those
• instruction-following, multilingual, and speed ratings omitted — checked 2026-07-20 (live browser fetch of both the model card and the news announcement). instruction-following/multilingual: neither primary source makes a claim on either dimension. speed: the model card has a 'Speed' tab (alongside Performance/Modalities) but its content renders as a chart, not extractable text — no textual claim to cite, so left out rather than guessed from a chart. Multilingual specifically may be out of scope by design (Lean 4 is a formal, not natural, language, and this model's entire training focus is proof engineering — but that's an inference, not a sourced claim, so it's left unrated rather than assumed)
• All benchmark data is vendor-reported (Mistral's own announcement); no independent third-party leaderboard exists for miniF2F, PutnamBench, FATE-H/X, or FLTEval — too niche for third-party benchmark sites to have picked up yet
• Active-parameter count is unresolved between Mistral's own two primary sources (6B vs 6.5B) — see architecture notes
• Free ($0) pricing is under Mistral's experimental 'Labs' branding — Mistral's own Labs policy discloses a 2-week deprecation notice and explicitly advises against production use; no scheduled future price exists, but the model/endpoint itself could be removed on short notice (docs.mistral.ai/models/labs)
• No independent corroboration of the 256k context window (single-source, model card only)

Citations

• Leanstral 1.5 model card (Mistral AI docs)
• Leanstral 1.5: Proof Abundance for All (Mistral AI announcement)
• Mistral Labs — model lifecycle policy (Mistral AI docs) — Source for the 2-week deprecation-notice and 'not recommended for production' disclosures — applies to all Labs-tier models generically, not Leanstral-specific.