FrontierSeptember 12, 2026via MarkTechPost
Cognition Releases SWE-2: A Kimi K3 Post-Trained Coding Model That Matches Fable 5.1 on FrontierCode at 64% Lower Cost
Why it matters
A post-trained coding specialist matches frontier performance on a major benchmark while dramatically undercutting cost-per-inference. This is capability parity through training efficiency, not raw model scale—a significant signal for the next wave of specialist models.
Key signals
- SWE-2 scores 50.0% on FrontierCode 1.1 Main
- Within 1 point of Fable 5.1 (51.0%)
- 64% lower cost than Fable 5.1
- Post-trained on Kimi K3 (2.8T-parameter open model from Moonshot AI)
- Cognition released SWE-2 as a specialized coding model
- Uses reinforcement learning for post-training
The hook
50.0% on FrontierCode, within 1 point of Fable 5.1—at 64% lower cost. Cognition's SWE-2 just reset coding-model economics.
Cognition, the company behind the Devin coding agent, has released SWE-2, its most capable coding model to date. SWE-2 is post-trained with reinforcement learning from Kimi K3, Moonshot AI’s 2.8T-parameter open model. Cognition reports a score of 50.0% on FrontierCode 1.1 Main, within 1 point of Fab…