Leanstral 1.5 Open-Sourced for Lean 4
sophiamyang · x · 2026-07-04
Sophia Yang releases Leanstral 1.5, a 119B (6B activation) open-source model designed for Lean 4 formal theorem proving. It achieves SOTA on miniF2F with 100%, PutnamBench with 587/672 (approx. $4/problem), FATE-H with 87%, and FATE-X with 34%, while discovering 5 previously unknown bugs in real open-source projects. It is licensed under Apache-2.0, with weights hosted on HuggingFace and a free API provided.
Related event: Mistral Releases Leanstral: Open-Source AI Agent for Lean 4(5 posts)→
More from Models
- Google’s year-long pause in new base-model pretraining draws sharp criticism — teortaxesTex · 2026-07-22
- Poolside’s Laguna S 2.1 launches as a 118B open-weight MoE with 1M context — NVIDIAAI · 2026-07-22
- Current setup is 8,192 input tokens and 2,048 output tokens, with 8k/512 next — TheZachMueller · 2026-07-22
- Kimi K3 feels slower than K2.7, but stronger on long coding jobs and refactoring — Far-Presence2711 · 2026-07-22
- Poolside launches Laguna S 2.1 with 118B parameters and 8B active per token — Madisonkanna · 2026-07-22
- What are the best models to run on 48 GB of VRAM with two RTX 3090s? — ludos1978 · 2026-07-22