Cognition SWE-2 First Test – Is THIS a Better Kimi K3?
- Source
- youtube.com
- Author
- Bijan Bowen
- Date
Context
A coding 's output can look promising while still leaving the intended workflow unfinished. In an informal test of Cognition's SWE-2, the reviewer asked the model for a printable engine design to fit inside a radio-controlled car. The generated STL files, which describe surfaces for 3D printing, would not open in the software used to prepare them for the printer.
The reviewer instead inspected the model's Blender scene and called the design "not bad" and decent enough, with a few printability issues, while noting the parts were never examined in a slicer. That assessment left print readiness unverified. The cause of the import failure was also unresolved. The distinction matters when evaluating Cognition's coding model: a reported visual assessment of the design and a successfully prepared print are different outcomes.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
Checking sign-in…
Loading comments…






