Cognition releases SWE-1.7, scoring 42.3 on FrontierCode benchmark using reinforcement learning on a Kimi base

Cognition announces SWE-1.7 which was trained from Kimi K2.7 base performance looks pretty incredible https://twitter.com/cognition/status/2074882968770728416

Benchmark scores are exciting but more importantly we are seeing incredible results so far using this model in Devin! Try it out: https://twitter.com/cognition/status/2074882968770728416

the proliferation of independent models is really picking up speed! very impressive work https://twitter.com/cognition/status/2074882968770728416

And Cog too?? 1000t/s and tops frontier code ! Busy busy week!! https://twitter.com/cognition/status/2074882968770728416

Incredible model release from Lux family co @cognition Maybe the most effective and efficient model now competive with frontier models but at fraction of cost....🤯 https://twitter.com/cognition/status/2074882968770728416

🚨COGNITION DROPS SWE-1.7 “Frontier intelligence at a fraction of the cost.” >trained on a Chinese base model (Kimi K2.7) >near Opus 4.8 / GPT-5.5 on benchmarks >1000 TPS via Cerebras “RL can push capabilities much further than previously believed.”… Show more