CognitionSWE
SWE-1.6
SWE-1.6 starts a new post-training run from the same pretrained base as SWE-1.5. Training targets unnecessary reasoning, repeated actions and slow tool use.
What the release provides
The model learns parallel tool use and uses a length penalty to reduce unproductive work. The launch offered standard serving through Fireworks and Fast serving through Cerebras with the same model intelligence. Fast is a service option, not another base model.Cognition: SWE-1.6
How to use it well
Practical guidance. Compare complete task results and response time under the intended workload. A published token-generation speed does not include tool execution or review time.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Provider product |
| Fast option | The CLI documentation includes the swe-1-6-fast selectionDevin CLI: Models |
| Current listing | Retained in the current Devin model catalogDevin: Current model catalog |
Related models
Kimi K2.7-based coding model trained to summarize its working context and continue long software-engineering tasks.
Launch date not confirmedProvider product