Anthropic open-sources proofs that identify which AI model answered
FlashProve can verify inference from committed model weights without revealing them, but the system remains an expensive research prototype.
Tags
Research
EngiWorld shows frontier agents still struggle with engineering software
The benchmark covers CAD, simulation, manufacturing, BIM, electronics and 3D workflows, with the strongest tested model reaching only 44.3 points.
Claude formalized Fermat's Last Theorem in Lean
Anthropic says the model worked largely autonomously for 11 days, turning a famous proof into machine-checkable code.
