Cognition says SWE-2 Flash runs three times faster than SWE-2 at one fifth the price, and it is free through November 30 in Devin Desktop and Devin CLI. Use the free window to measure it against Claude Haiku 5.5 on your own tasks, because Cognition's own table shows a mixed picture.
Cognition released SWE-2 Flash on November 4, 2026, and says it is available in Devin Desktop and Devin CLI at one fifth the price and three times the speed of SWE-2. Its generation speed demo shows 200 tokens per second for SWE-2 Flash against 60 tokens per second for SWE-2, replayed on a Terminal-Bench 4.0 task. For anyone running coding agents all day, that combination is worth a real test.
The timing is the practical tip. Cognition says SWE-2 Flash is free through November 30 in Devin Desktop and Devin CLI, so you can run it on live work without a budget approval. Treat the free window as a measurement period with a deadline, and decide what you will measure before you start, not after.
Read Cognition's own comparison table carefully, because it does not show a clean win. The announcement compares SWE-2 Flash with Claude Haiku 5.5 as follows: Terminal-Bench 2.1 89.2 against 94.4, SWE-Bench Pro v2 Hard 89.9 against 90.2, DeepSWE 1.1 60.1 against 59.3, FrontierCode 2.0 24.0 against 27.7, ExploitBench 34.8 against 38.0, and SWE-Atlas QA 50.0 against 56.5. SWE-2 Flash trails on most rows and edges ahead on one. The case for it is speed and price on routine work, not a claim to beat Haiku 5.5 everywhere.
One constraint to plan around: Cognition says it is not sharing SWE-2 Flash weights publicly yet because of cyber misuse risk, and that it is building a program to share them with trusted industry partners and research labs. That means you can use the model only through Cognition's tools for now, so check how it fits your workflow before you build habits around it.
Try it today
- Pick 10 to 20 real tickets from your backlog that are routine, such as small bug fixes, test additions and refactors, and keep the same tasks for every model you compare.
- Run them through SWE-2 Flash in Devin Desktop or Devin CLI before November 30, and run the same tasks through Claude Haiku 5.5.
- Record time to finish, cost per completed task, and how many results needed human rework, rather than relying on benchmark scores.
- Route the routine work that passes to SWE-2 Flash, and keep harder tasks on your stronger model.
- Set a calendar reminder for November 29 to decide whether to keep paying for it once the free period ends.