Tasks and environments for AI development
Collinear AI builds training tasks, environments and evaluations for AI developers. Its work focuses on verifiable tasks that can supply feedback during model development, including coding and other tool-using workflows. These are services and artifacts supplied by the company, rather than an independent certification of a customer's model. [1]
Collinear released the public CWE-bench v2 set on November 3, 2026. The benchmark tests whether agents can repair security weaknesses without breaking legitimate behavior, using executable scoring. The launch distinguishes the public set from held-out evaluation tasks and directs users through a gated dataset to obtain the public files. [2]
Sources
- AI Simulation Lab for Training Data & RL
Collinear AI · retrieved Nov 3, 2026
- CWE-bench v2: More coverage, quality, and transparency
Hugging Face · Nov 3, 2026