Cognition launches SWE-2 Flash coding model, beats US open-weight models 3x its size
EricBuess · x · 2026-11-05
Cognition announced SWE-2 Flash, a new fast and efficient coding model it says outperforms all American open-weight models on leading evals — even ones 3x larger. To balance transparency and cyber risk, the company is building a program to securely share weights with partners and researchers. SWE-2 and SWE-2 Flash are free in Devin Desktop and Devin CLI through November 30.
Related event: Cognition launches SWE-2 Flash, a faster, cheaper 276B MoE coding model(6 posts)→
More from coding & agent
- DeepSeek's open stack mapped: 10 building blocks from V4.1-Flash to 3FS — and what's missing — TheTuringPost · 2026-11-05
- LukeW: no design system survives contact with the product — agents fill the gap — LukeW · 2026-11-05
- US banks split on AI agents: big banks block them, small banks build agent-native accounts — rohanpaul_ai · 2026-11-05
- Hackathon project PMO gives AI agents permanent memory, cutting tokens by half — AIandDesign · 2026-11-05
- Dev observation: most apps are just becoming databases for agents — majidmanzarpour · 2026-11-05
- He built an in-app video studio that renders, exports and auto-posts to socials — tomjohndesign · 2026-11-05