Mistral AI · 2026-06-30 · notable
Leanstral 1.5 — Mistral's updated Lean 4 formal-proof model
Leanstral 1.5 updates Mistral's Lean 4 theorem-proving model with a 119B-parameter MoE, 6.5B active weights, and a 256K context window. Free to try in Mistral Labs playground under model id labs-leanstral-1-5.
Mistral updates its Lean 4 formal-proof model with a wider context window and a free Labs playground slot.
Key specs
| Parameters | 119B |
|---|---|
| Active params | 6.5B |
| Context window | 256K |
| Price | Free (Labs) |
What is it?
Leanstral 1.5 is a domain-tuned coding model that writes Lean 4 proofs and rewrites informal math into machine-checkable statements (autoformalization). It succeeds the March 2026 original Leanstral and now runs under model id labs-leanstral-1-5 on Mistral's Labs endpoint.
How does it work?
Under the hood is a mixture-of-experts network with 119B total parameters and 6.5B activated per token, giving big-model recall at small-model compute. The 1.5 update pushes context to 256K tokens so full proof files, Mathlib imports, and prior tactics can sit in the same prompt.
Why does it matter?
Automated proof search is one of the few coding tasks with a hard ground-truth checker (the Lean kernel), so a stronger formal-proof model reduces the busywork in verified software and formal mathematics. Free access in Labs lets researchers benchmark it against DeepSeek-Prover, Kimina, and Goedel-Prover without a paid tier.
Who is it for?
Formal-methods researchers, Lean 4 users, and verified-software teams.
Try it
Model id: labs-leanstral-1-5 on Mistral's API (free tier in Labs).