Cognition Ships SWE-2, Its First Model With Effort Levels, Pushing the Pareto Curve in a Single RL Run
andrew_n_carr · x · 2026-09-11
Cognition released SWE-2, its first coding model to support effort levels, enabled by significant improvements to its length penalty recipe.
Key result: a single RL run pushes the Pareto curve while preserving its shape — medium effort becomes both cheaper and smarter, while max effort learns to spend more tokens and turns to hit the highest scores. The accompanying chart drew praise from researchers like Andrew Carr.
Related event: Cognition Launches SWE-2 and Raises $2 Billion(5 posts)→
More from coding & agent
- Claude Managed Agents adds session viewer: attach ant CLI to live agent sessions — ClaudeDevs · 2026-09-11
- webmcp is a bridge, not the end state — the web will go truly agentic — EdenEmarco177 · 2026-09-11
- Tencent open-sources WeKnora, a RAG knowledge platform with 22k GitHub stars — abhishek__AI · 2026-09-11
- Tencent open-sources WeKnora, a RAG platform turning scattered docs into reasoning knowledge bases — abhishek__AI · 2026-09-11
- One Pokédex, four model setups: hands-on comparison of GPT-6 Astra and DeepSeek V4.1 Flash — kevinkern · 2026-09-11
- Agent Arena launches leaderboard grading models on millions of real agentic tasks — arena · 2026-09-11