Jev vs. GLM at chess (X)

Play the video in the post below

Hassan pitted Jev against GLM 5.3 in a full game of chess, with each model choosing moves for one side. GLM 5.3 won by checkmate in 29 moves. On speed and cost, Jev returned a move in about 0.3 seconds for under $0.0001, while GLM 5.3 took roughly 5.8 seconds and about $0.008 per move. The full game cost 24 cents.

Hassan’s takeaway is to route by task: fast, well-defined classification to a specialized model like Jev, and classifications that need reasoning or lookahead to an LLM like GLM 5.3. In a real pipeline, he suggests a hybrid where Jev handles the easy calls and an open model handles the hard ones.

Open on X