Cognition says SWE-2 Flash, its new fast and efficient coding model, is free through November 30 for Pro, Max and Teams users in Devin Desktop and Devin CLI. The company says it costs one-fifth as much as its larger SWE-2 model and runs three times faster, according to its launch blog.
What the model is built on
SWE-2 Flash starts from Inkling-Small, a model from Thinking Machines with 276B total and 12B active parameters. Cognition says it added a mid-training stage before reinforcement learning for the first time, training on several trillion tokens of scientific articles, blogs, code and other sources. The highest-quality data was placed last in the sequence.
Image credit: @cognition on X
Cognition says most of its reinforcement learning gains came from better and more varied tasks rather than a changed recipe. Its team audited task instructions, environment setup and verifiers.
Image credit: @cognition on X
Benchmarks
The table below is Cognition's own, comparing SWE-2 Flash with the Inkling-Small base it started from.
| Benchmark | Inkling-Small base | SWE-2 Flash |
|---|---|---|
| Terminal-Bench 4.0 | 0.9% | 20.0% |
| FrontierCode 2.0 | 3.2% | 24.0% |
| ExploitBench | 11.3% | 34.8% |
The same table includes rivals, and SWE-2 Flash does not lead every row. On DeepSWE 1.1 it scores 60.1 against 77.0 for DeepSeek V4.1 Flash, and on Terminal-Bench 4.0 it trails GLM 5.3 Flash (41.8) and DeepSeek V4.1 Flash (42.3). Cognition says it outperforms all American open-weight models on leading evals, including ones about three times its size.
Image credit: @cognition on X
Cognition also published a replay of one Terminal-Bench 4.0 task at a simulated 200 tokens per second for SWE-2 Flash against 60 for SWE-2. That is one task, not a broad measure.
Price and access
SWE-2 Flash is available in Devin Desktop and Devin CLI. The launch post makes it free until the end of the month for Pro, Max and Teams users. Cognition's Devin Desktop page lists the plans: Pro at $20 per month, Max at $200 per month, and Teams at $80 per month plus $40 per full seat.
Devin Desktop is the new name for Windsurf. The Devin Desktop announcement says the change arrives as a standard over-the-air update and that plans and pricing stay the same.
Weights stay private for now
Cognition is not releasing the SWE-2 Flash weights publicly. It cites cyber risk, pointing to a recent surge in cyber attacks, and says it is building a program to share weights securely with trusted industry partners and research labs. The company says it will share more in a few weeks.
Our read: for teams already paying for Devin, the free window is the cheapest way to test whether a smaller model can handle routine work such as code review and documentation updates. Teams that need the weights should wait for the partner program details.
For more on running several coding agents in one app, see our earlier report on Kimi Code Desktop.
0 comments
No approved comments yet. You can start the conversation.
Leave a comment