Cognition has released SWE-2, a new coding model post-trained using reinforcement learning from Moonshot AI's Kimi K3 model. This new model reportedly matches Fable 5.1's performance on the FrontierCode benchmark but at a significantly lower cost. SWE-2 also introduces selectable reasoning-effort levels, trained in a single reinforcement learning run, and demonstrates improved efficiency and reduced detours on coding tasks compared to its predecessor, SWE-1.7. AI
IMPACT Sets a new cost-performance benchmark for coding models, potentially influencing future development and deployment strategies.
RANK_REASON New model release from a frontier lab (Cognition) with performance benchmarks. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- cognition
- Devin
- Fable 5.1
- FrontierCode
- GPT-6 Astra
- Kimi K3
- Moonshot AI
- SWE-1.7
- Terminal-Bench 2.1
- Terminal-Bench 4
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
