Leanstral 1.5: Open-Source AI Model Achieves State-of-the-Art Results in Formal Verification and Proof Engineering
By
programLyrique
Summary
Leanstral 1.5 is a free, Apache-2.0 licensed AI model with 6B active parameters focused on formal verification and proof engineering. It achieves state-of-the-art results across multiple benchmarks including saturating miniF2F, solving 587/672 PutnamBench problems, and scoring 87% on FATE-H and 34% on FATE-X. Trained using mid-training, supervised fine-tuning, and reinforcement learning with CISPO, it excels at agentic proof engineering and real-world code verification, uncovering 5 previously unknown bugs across 57 repositories. The model is fully open-sourced via Hugging Face and a free API.
Source
Hacker NewsLeanstral 1.5: Open-Source AI Model Achieves State-of-the-Art Results in Formal Verification and Proof Engineeringmistral.aiKey quotes
· 4 pulledLeanstral 1.5, a free Apache-2.0 licensed model with 6B active parameters, delivers a major performance upgrade in formal verification
saturating miniF2F, solving 587/672 PutnamBench problems, and achieving state-of-the-art results on FATE-H (87%) and FATE-X (34%)
excels in agentic proof engineering and real-world code verification, uncovering 5 previously unknown bugs across 57 repositories tested
Fully open-sourced and available via Hugging Face and a free API
You might also wanna read
Phi-4 Reasoning: Small Open-Weight AI Models with Strong Math and Science Capabilities
Phi-4 Reasoning is a small open-weight language model (3.8B/14B parameters) that delivers powerful reasoning capabilities for math, science,
Product Hunt·3mo ago
MerLean-Prover: A Recursive Agent Harness for Lean 4 Theorem Proving Outperforms Baselines
MerLean-Prover is an end-to-end Lean4 theorem prover that replaces 'sorry' declarations with kernel-checkable proofs using three agent types
arxiv.org·1mo ago
GLM-5.2 Open-Weight Model Outperforms Opus 4.8 on AI-Resistant Backend Test
The article presents a detailed technical comparison between GLM-5.2 (open-weight model) and Opus 4.8, demonstrating that GLM-5.2 outperform
southbridge.ai·10d ago
Jatevo.ai: A Multi-Model LLM Inference Load Balancer
Jatevo.ai is an OpenAI-compatible inference cloud that aggregates multiple LLM providers, GPU pools, and deployment lanes into a single gate
Arcee AI Launches Trinity-Large-Thinking: Open-Source AI Model Matching Opus 4.6 Performance at 96% Lower Cost
Arcee AI has launched Trinity-Large-Thinking, an open-source AI model that claims to match the performance of OpenAI's Opus 4.6 while being
Product Hunt·3mo ago
Unsloth: Open-Source Platform for Local AI Model Training and Inference
Unsloth is an open-source platform that enables users to run and train AI models and large language models (LLMs) locally on their own hardw
Product Hunt·3mo ago
Comments
Sign in to join the conversation.
No comments yet. Be the first.