Cognition launches SWE-2 Flash coding model, achieving 5x lower cost and 3x faster speed based on Inkling-Small
SWE-2 Flash is based on @thinkymachines Inkling-Small, an American model with 276B total and 12B active parameters. The small size, combi...
AI BriefingCognition launched SWE-2 Flash, based on Inkling-Small, a US model from thinkymachines, with 276B total parameters and 12B active parameters. Combined with a highly optimized inference stack, it is priced 5x lower and runs 3x faster than SWE-2.
Also reported by 2 other sources
- Cognition Releases SWE-2 Flash Coding Model, Raising Terminal-Bench 4.0 from 0.9% to 20.0%Cognition Models / Devin Blog · 12 days ago
- Cognition releases SWE-2 Flash coding modelCognition · 12 days ago
Source: Cognition · x.com
Event follow-up1
View all event coverage- Cognition Releases SWE-2 Flash Coding Model, Raising Terminal-Bench 4.0 from 0.9% to 20.0%Cognition Models / Devin Blog · 12 days ago