Devin users who want a cheaper, fast coding model for long tasks, though SWE-2 has since replaced it as the newer option.
Kenning's take
What it does
SWE-1.7 is a software engineering model from Cognition, trained from a Kimi K2.7 base with further reinforcement learning. It is tuned for long, asynchronous coding tasks and can summarise its own working state to keep going past its context window. It is available in Devin Web, Desktop and CLI, including a faster Lightning version served on Cerebras, and has since been followed by SWE-2.
- Trained from a Kimi K2.7 base with Cognition's RL pipeline
- Built for long, asynchronous coding tasks
- Self-compaction to work past the context window
- SWE-1.7 Lightning runs on Cerebras at about 1,000 tokens a second
- Medium and Max effort versions
Who it suits. Developers using Devin who want a cheaper, fast coding model.
Key facts
- Category
- AI models
- Made by
- Cognition
- Released
- 8 July 2026
- How to get it
- in Devin Web, Devin Desktop and Devin CLI
- Works on
- Devin Web, Devin Desktop, Devin CLI
- API price
- enterprise rate for SWE-1.7 Medium listed at $0.50 per million input tokens and $2.50 per million output; SWE-1.7 Lightning Medium $2.50 and $12.50
Pricing
| Plan | Price | Billed |
|---|---|---|
| Pro | $20 | month |
| Max | $200 | month |
FAQ
When was SWE-1.7 released?
Cognition launched SWE-1.7 on 8 July 2026.
What is SWE-1.7 Lightning?
A faster version of SWE-1.7 served on Cerebras, with the same intelligence and lower latency.
Is SWE-1.7 still Cognition's newest model?
No. Cognition's docs call it the previous flagship, after SWE-2.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
ACE-Step 1.5
Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.
Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.
Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.