{"slug": "openai-says-agents-now-finish-nearly-half-of-2-to-4-hour-research-tasks-on-their", "title": "OpenAI says agents now finish nearly half of 2-to-4-hour research tasks on their own", "summary": "OpenAI said in a November 18 report that its coding agents now complete nearly half of research tasks estimated to take a human two to four hours, roughly double the success rate on such tasks since March. OpenAI, led by co-founder and CEO Sam Altman, also argued that AI labs should be required to track and publicly report indicators of recursive self-improvement.", "body_md": "# OpenAI says agents now finish nearly half of 2-to-4-hour research tasks on their own\n\n**In its November 18th report, OpenAI says agent success doubled on longer research tasks since March and calls for labs to be required to report recursive self-improvement indicators.**\n\n        By [Ryan Merket](https://runtimewire.com/author/ryan-merket)\n        · Published \n\nOpenAI, led by co-founder and CEO Sam Altman (@sama), said on November 18th that its coding agents now complete nearly half of research tasks estimated to take a human two to four hours, and argued that AI labs should be required to track and publicly report indicators of recursive self-improvement.…", "url": "https://wpnews.pro/news/openai-says-agents-now-finish-nearly-half-of-2-to-4-hour-research-tasks-on-their", "canonical_source": "https://runtimewire.com/article/openai-says-agents-now-finish-nearly-half-of-2-to-4-hour-research-tasks-on-their", "published_at": "2026-11-18 23:14:51+00:00", "updated_at": "2026-11-18 23:42:18.322235+00:00", "lang": "en", "topics": ["ai-agents", "ai-safety", "ai-policy", "ai-research"], "entities": ["OpenAI", "Sam Altman"], "also_reported_by": [], "takeaway": "OpenAI reports its coding agents now finish nearly half of research tasks estimated at two to four human hours, up from roughly a quarter in March.", "recommendation": {"advice": "Track agent success rates on tasks bucketed by human completion time, such as the 2-to-4-hour band, to detect capability gains that aggregate benchmarks hide.", "playbook": "evals-testing", "playbook_url": "https://wpnews.pro/playbooks/evals-testing"}, "alternates": {"html": "https://wpnews.pro/news/openai-says-agents-now-finish-nearly-half-of-2-to-4-hour-research-tasks-on-their", "markdown": "https://wpnews.pro/news/openai-says-agents-now-finish-nearly-half-of-2-to-4-hour-research-tasks-on-their.md", "text": "https://wpnews.pro/news/openai-says-agents-now-finish-nearly-half-of-2-to-4-hour-research-tasks-on-their.txt", "jsonld": "https://wpnews.pro/news/openai-says-agents-now-finish-nearly-half-of-2-to-4-hour-research-tasks-on-their.jsonld"}}