Cognition launches SWE-2 Flash coding model, post-trained on Inkling-small
mishig25 · x · 2026-11-05
Cognition introduced SWE-2 Flash, a new fast and efficient coding model that reportedly outperforms all American open-weight models, even ones 3x larger. To balance transparency and cyber risks, the company is building a program to securely share weights with partners and researchers. Soumith Chintala notes it was post-trained on top of Inkling-small.
Related event: Cognition Releases SWE-2 Flash, Beating Larger US Open Models(3 posts)→
More from coding & agent
- YC-affiliated Weave open-sources agent that enforces Markdown code rules on every PR — ycombinator · 2026-11-05
- Why AI agents break in enterprise finance: 25% of invoices lack a PO — vasuman · 2026-11-05
- Twilio shows voice models + phone calls + MMS with OpenAI GPT-Live 1 — craigsdennis · 2026-11-05
- One table to tell InstaBox, boxd and Entire apart: three layers of the agentic stack — ChrisUniverse · 2026-11-05
- Priors adds best-quote routing for stock tokens via Rialto, with MCP access for agents — econoar · 2026-11-05
- Merge's eval claims its MCP connectors beat Microsoft's native servers on tool count and reliability — shensi · 2026-11-05