Demonstrations documented: Andon Labs drone tracking, multi-agent election simulation, 3D environment creation
Cognition Adds GPT-6 Astra to Devin, Releases SWE-2 and Devin Voice
The integration extends AI-assisted software development toward automated testing and verification, reducing the manual review burden on engineers. Cognition's use of GPT-6 Astra also marks one of the first disclosed product deployments of OpenAI's newest flagship model.
The full picture
Cognition integrated OpenAI's GPT-6 Astra across its Devin product lineup, including the core cloud agent, CLI, and desktop products, with the stated goal of letting Devin test its own output and produce evidence that software functions as expected, reducing how much code engineers must review manually. In a demo, Devin used Astra to test an iPhone game called Otter Run, returning a simulator recording and a testing report. A bug-fix workflow was also described in which Devin receives a screenshot of a bug, fixes the code, and returns a screenshot confirming the fix. Cognition said it expects engineers will "manually look at less code and end up shipping more at the end of the day" as a result of the GPT-6 integration. Separately, Cognition released SWE-2, a coding model based on Kimi K3, and Devin Voice, which pairs GPT-Live with Devin so users can verbally describe a coding task before handing it off to the agent to build.
How it developed
SWE-2 and Devin Voice launch reported
MazeBench results published: Astra scored 14% without Python, above Fable 5.1's 2% and Sol's 13% with code
@anshuc used Astra to build a game-ready afterlight courtyard in Blender in approximately 45 minutes
Sources
Related
- Split fromCognition's funding round for Devin
Want this in your inbox?
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free