Play the video in the post belowPost on X
- Views
- 559,425
- Likes
- 2,343
- Reposts
- 150
Read from X on . Counts change daily.
AI/ML API set up TypeSafe’s Jev V13 to play 5+0 blitz chess (five minutes per side, no increment) against two frontier language models, Fable 5.1 and GPT-6 Astra, with every move made through one API call. The post introduces Jev as a model that does not chat, explain or write code, and only makes decisions.
Against Fable 5.1, the post says Fable outplayed Jev, going up 16 points of material by move 29 and promoting a second queen. But Fable spent 6 to 15 seconds per move on analysis against Jev’s roughly 2.6 seconds, and ran out of time and lost. Against GPT-6 Astra the result went the other way: Astra did not play for material and mated Jev in 18 moves with Qe1#, with 2:27 left on its clock.



