OpenAI's Astra model solves decade-old math problems
OpenAI used an internal version of its next model, Astra, to solve ten math problems with no progress for at least a decade, spending under $2,000 per problem at GPT-5.6 Sol token prices. Results include Lean 4 formalizations and a paper.
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- DeepMind panel discusses generative media SOTA and human eval
- Reddit users share useful MCP servers for Claude
- Why embodied AI hits an edge AI wall requiring new math
- HTMX CEO mandates 'No AI Fridays' to counter LLM cognitive debt
- OpenAI product lead Tara Seshan discusses persistent AI coworkers