Kimina-Prover: structured formal reasoning and reinforcement learning

What changed

Used mixed informal/formal cold-start examples, a structured reasoning pattern with Lean snippets, and verifier-reward RL with format/gradient filtering for whole-proof generation.

Scientific scope

Used mixed informal/formal cold-start examples, a structured reasoning pattern with Lean snippets, and verifier-reward RL with format/gradient filtering for whole-proof generation. Inherits Qwen2.5, Kimi RL, synthetic teacher data and Lean. Final-proof rewards do not certify faithful hidden reasoning or statement translation. High pass@8192 results use corrected benchmark statements and substantial sampling; front collective is not40individual front authors.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Inherits Qwen2.5, Kimi RL, synthetic teacher data and Lean. Final-proof rewards do not certify faithful hidden reasoning or statement translation. High pass@8192 results use corrected benchmark statements and substantial sampling; front collective is not40individual front authors.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Kimina-Prover Preview: Towards Large Formal Reasoning Models with Reinforcement Learning

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction