HeadsUpAI

Cognition Releases SWE-1.6 with Optimized Agent UX and Unmatched Speed

· Updated

Cognition, an applied AI lab building software engineering agents, released SWE-1.6 for general availability in the Windsurf IDE. While it matches previous benchmarks, the model is post-trained (refined after initial training) from scratch to improve Model UX—the behavioral traits that determine how smoothly an agent completes multi-turn tasks.

This update shifts focus from raw intelligence to agentic efficiency. By introducing a length penalty during training, the model is forced to find shorter paths to solutions, reducing looping behavior (repetitive reasoning cycles) and overthinking. It now favors parallel tool calls and specialized tools over generic terminal commands.

You can access SWE-1.6 today in Windsurf, where it is free for all users for the next three months. A free tier served by Fireworks runs at 200 tokens per second (speed of text generation), while a fast tier for paying users, powered by Cerebras, reaches 950 tokens per second.

Cognition
Cognition
@cognition
X

We’re releasing SWE-1.6, our best model in both intelligence & model UX. SWE-1.6 matches our Preview model on SWE-Bench Pro while dramatically improving on various behavioral axes. It’s available today in Windsurf in two modes: free tier (200 tok/s) and fast tier (950 tok/s). https://t.co/JHwMLsaudO

55retweets627likes
View on X

Share this update