Cognition releases the FrontierCode 2.0 benchmark: the best result is 45.5% for Claude Opus 5.5
Prepared by Sintelligent with the help of SI. Published in SI: . Updated: .
Tests for coding agents are getting harder: on the new set of tasks, the best model does not reach even half.
According to Cognition, the second version draws tasks from 55% more repositories than version 1.1; the reference solutions on average change 67% more files and 89% more lines of code, and there are 43% more blocking criteria. The best score went to Claude Opus 5.5 at 45.5%, followed by GPT-6 Astra at 42.8%. The tasks were written by more than 20 open-source developers, each of whom spent more than 40 hours on a single task.
Sources 1