OpenAI Astra generates Lean proofs for 10 math problems

// 1h agoNEWS

OpenAI Astra generates Lean proofs for 10 math problems

OpenAI’s Astra model produced formal Lean proofs for 10 mathematical problems that had sat open for years, resolving four of them via counterexamples. This breakthrough highlights the growing capability of AI models to perform machine-checkable formal verification in mathematics, while competitive Chinese models in the same week dramatically lowered the cost of running comparable reasoning tasks.

// ANALYSIS

Automated formal proof generation is transitioning from an academic experiment into a practical benchmark for advanced AI reasoning, though rapid price drops will quickly democratize access.

–Demonstrates AI's capacity to resolve long-standing open problems with machine-verifiable proof certificates in Lean.
–Discovering four counterexamples proves AI can identify structural flaws in existing mathematical conjectures rather than just validating known patterns.
–Simultaneous cost reductions from open and Chinese models indicate high-level reasoning and formal verification capabilities will quickly become affordable at scale.

// TAGS

openaiastraopenai-astraleanmathematicsformal-verificationai-reasoning

DISCOVERED

1h ago

2026-08-04

PUBLISHED

2h ago

2026-08-04

RELEVANCE

9/ 10

AUTHOR

jecrosbie

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

OPEN SOURCE23m ago

forecast_agents scores prediction markets using AI

forecast_agents is a free and open-source system powered by seven AI agents that evaluate prediction market questions on platforms like Kalshi and Polymarket. Rather than executing automated trades or accessing brokerage accounts, the multi-agent setup provides users with calculated probability scores to inform independent decision-making.

RESEARCH34m ago

SESA combines zero-data self-play with skill memory evolution

SESA (Self-Evolving Skill-Augmented Agent) introduces a novel framework that integrates procedural skill memory evolution into zero-data self-play search agents. By co-evolving task generators and procedural memory across expanding learning frontiers, SESA enables AI agents to autonomously pose, solve, and consolidate procedural skills without relying on predefined external training data.

RESEARCH34m ago

RecHarness automates recommender optimization with bandit routing

RecHarness is an open-source research framework that automates recommender system optimization by combining a multi-armed bandit router with an LLM code synthesis agent. The bandit router selects strategic modification directions while the agent generates executable code edits, using a jump-basin mechanism to escape local optima during long-horizon tuning.