startup
Surprise upset: GPT-5.5 beats Claude Fable 5 on brutal new Agents’ Last Exam benchmark
Source:
venturebeat.com 1 min read
Share
You are reading a summary. The full content is hosted on venturebeat.com.
UC Berkeley RDI and 300+ experts launched Agents’ Last Exam, a benchmark of long-horizon professional workflows across 55 industries with mostly deterministic grading and anti-contamination controls. GPT-5.5 via Codex leads the leaderboard at a 24.0% pass rate, underscoring that top models still perform poorly, with many scoring 0.0% on the hardest tier.
Read the full article on the original website
External link to venturebeat.com
Related Articles
startup
Scientists Warn a Popular Joint Supplement May Accelerate Your Risk of Cognitive Decline—Here’s What to Know
1 min read •
startup
South Korea’s Floundering Movie Business Turns to AI for Help
1 min read •
startup
Sources: Frank founder Charlie Javice, sentenced in September 2025 to 85 months for defrauding JPMorgan Chase, has been seeking a presidential pardon from Trump (Wall Street Journal)
1 min read •