Cognition launches SWE-2 Flash for Devin
The Inkling-Small-based coding model is available in Devin Desktop and CLI. Cognition reports roughly three times the generation speed of SWE-2 at one-fifth of its price.
Why it matters
For teams using Devin on recurring coding tasks, the trade-off is between cost, latency, and output quality. Try it on small, easy-to-review changes before handing it work that spans several parts of a codebase.
Start with tasks with a clear outcome, such as updating one test or fixing a reproducible bug. Run the same task several times and track how often the result passes tests without intervention. This gives a clearer picture of the cost of a usable change, including the time reviewers spend on it. Keep Cognition’s generation speed figures separate from the duration of the whole job. Setting up the environment, running tests, and reviewing a diff also take time. Measure the interval from task start to a reviewable result when deciding what work to assign.