TypeSafe's Jev Model Scores Mid-Pack Chess Elo at $0.0015 Per Game

TypeSafe's Jev is an unconventional AI model that accepts structured inputs with predefined label options and returns probabilistic classifications, rather than generating free-form text. A developer tested Jev on the LLM Chess benchmark by framing each move selection as a Choice query over all legal UCI moves, allowing it to participate in the same evaluation harness used for standard chat models. Jev achieved an Elo rating of approximately 243, placing it near mid-tier reasoning models like o4-mini-medium, while completing games in around 36 seconds at roughly $0.0015 each — far cheaper than its leaderboard neighbors. The model maintained a consistent 50% draw rate against a chess engine across three difficulty levels, with zero illegal moves or protocol failures, highlighting the reliability of its typed, constrained output format. TypeSafe itself does not pursue public benchmarks, so this represents an independent external evaluation showing Jev performs well on tasks involving selection from a fixed list of options.