Cognition Launches SWE-2 Flash: 3x Faster at One-Fifth the Price
winer666 · reddit · 2026-11-05
Cognition (the company behind Devin) announced SWE-2 Flash on Nov 4: a cheaper, faster sibling to SWE-2 at roughly 3x the speed and one-fifth the price, aimed at routine agent work — automations, code review, feature development, docs — rather than benchmark leadership.
Key details:
- 276B total / 12B active parameters, adapted from Thinking Machines Lab's Inkling-Small (full weights released July 30)
- Big gains over Inkling-Small but mixed results vs the larger SWE-2; positioned as a cost-and-speed alternative, not a replacement
- Weights withheld for now behind a secure sharing program; targeted cybersecurity training lifted ExploitBench from 11.3% to 34.8%, a capability Cognition worries could aid attackers too
- Free through Nov 30, 2026; available in Devin Desktop and CLI
Open questions: does speed + low price beat benchmark leadership for daily coding, and does withholding weights slow independent verification of safety claims? (Figures are Cognition's own.)
Related event: Cognition Launches SWE-2 Flash, Beating Larger US Open-Weight Models(5 posts)→
More from coding & agent
- Model gets stuck in a loop spamming "it failed!!!" — burning 140K tokens — CtrlAltDwayne · 2026-11-05
- AX to support migrating agentic workloads between local and production clusters — rakyll · 2026-11-05
- OpenAI upgrades Codex: marked-up images now pinpoint edits via API — OpenAIDevs · 2026-11-05
- Baseten Router cuts coding-task costs 72% with only 0.8pt pass-rate drop vs Opus 5.5 — baseten · 2026-11-05
- BrowserForge: open-source MCP server gives agents your real browser with human approval gates — mostafazog · 2026-11-05
- Free course teaches analytics engineers AI skills beyond Snowflake and dbt — Zachly · 2026-11-05