Cognition SWE-2 Flash
Quiet signalCognition’s fast coding model, 3x the speed of SWE-2
Launched 3d ago
Signal
20
Velocity
0
What this means
30%Vote pace is below typical for AI Agents.
Running at 30% of comparable launches.
Prediction
Top-5 finish probability
2%
today
Projected end-of-day votes
3range 2–4
Trajectory
stable
Vote pace holding steady.
Speed vs peers
0.3×
11 AI Agents launches
About
SWE-2 Flash is Cognition’s new coding model, built on Thinking Machines’ Inkling-Small with 276B total and 12B active parameters. Per Cognition, it generates about 200 tokens per second, 3x the speed of SWE-2 at one-fifth the price, and scores 20.0% on Terminal-Bench 4.0, up from 0.9% for the base model. Try it in Devin Desktop and Devin CLI, free through 11/30. Weights are not public yet.
AI Summary
Cognition SWE-2 Flash is an advanced coding model with 276B total and 12B active parameters, achieving a generation speed of 200 tokens per second. It offers a significant performance improvement over SWE-2, scoring 20.0% on Terminal-Bench 4.0, and is available for testing in Devin Desktop and CLI until November 30.
Attention & discussion velocity
Performance
Velocity0
Vote pace vs average
Momentum0
Sustained over 6h
Virality0
Spread × engagement
Engagement—
not measured
Signal sources
2 of 6 active- Launch tracking7 readings since we picked it up
- AI analysisAI Agents
- Comment intelligencethread not yet analysed
- Outside discussionnot checked — domain is a storefront or shared
- Revenue signalsno match on this domain
- Editorial scoringbelow the signal threshold for this pass
A source that found nothing is a measurement. A source that has not run is a gap. Neither means the launch lacks the thing.