Benchmarks & research
Henrique Macedo pits Jev against Clef in 300 chess games, using the same legal-move choices and Stockfish to grade play.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
Clef won the pairing in the maker’s report, though Jev was faster and cheaper. Most games ended in draws through repeated positions. With only one prompt and a policy that always picks the top-scored move, the comparison cannot establish which model is generally better at chess.
The game code lists the legal moves from a position, asks the model to choose, then checks and applies its move. That separation is the part you can reuse: the AI chooses only among legal actions. A confident choice is still no promise of strong chess play.