Built on Thinking Machines' Inkling-Small, the model is free for paid Devin users through November 30th while its weights stay behind a gated program.
By [Ryan Merket](https://runtimewire.com/author/ryan-merket)
· Published
Primary source: [X · @cognition](https://x.com/cognition/status/2118069614521241739?a=3AyxUBZV92-7A53PVUY&s=46)
Why it matters #
SWE-2 Flash tests whether a coding-agent company can cut its cost per task by post-training a small American open-weight base instead of renting frontier capacity for every job. Cognition says the model runs at one-fifth the price and three times the speed of SWE-2. Its gated weight program also sets a precedent for controlled transparency: partners and researchers get access, the public does not, and Cognition keeps the distribution switch.
Cognition, the company behind Devin, launched SWE-2 Flash on November 4th, 2026, positioning the model as the cheaper and faster option for routine engineering work inside Devin Desktop and Devin CLI. Co-founder and chief executive Scott Wu (@ScottWu46) leads the company with co-founders Steven Hao and Walden Yan, a founding team drawn from competitive programming: all three are International Olympiad in Informatics gold medalists, according to published accounts of the company.…
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.