Co-Design Bench testing reveals that coding agents achieve full compliance on only about one-quarter of tasks.
Among ten tested models, the top performer, GPT-6 Astra, met approximately 70% of task requirements on average but achieved complete correctness in only 26.4% of tasks. Claude Opus 5 and Qwen3.8-Max followed closely, while Grok 4.6 finished last.
Return to original text
A short introduction is retained here; the full text from the source website is not reposted. For specific wording and subsequent updates, please refer to the original article.
Open AIHOT original