Cognition has released SWE-1.6, their latest coding model, available today in Windsurf on two tiers: free (200 tok/s) and fast (950 tok/s).
The headline numbers: SWE-1.6 matches the previous Preview model on SWE-Bench Pro while using 40% fewer turns to get there. The team describes the improvement as progress on “intelligence per token” - the model was post-trained from scratch with the same pretraining base, but does more with less output.
At 950 tok/s on the fast tier, it’s currently the fastest model they’ve shipped.

Collections

Cognition has released SWE-1.6, their latest coding model, now available in Windsurf on free (200 tok/s) and fast (950 tok/s) tiers. The model matches the previous Preview model on SWE-Bench Pro benchmarks while requiring 40% fewer turns, reflecting improvements in 'intelligence per token'. Post-trained from scratch on the same pretraining base, SWE-1.6 is currently the fastest model Cognition has shipped at 950 tok/s on the fast tier.

Cognition releases SWE-1.6 with 40% fewer turns and 950 tok/s on fast tier