Mercor: GPT-6.2 Astra debuts at #1 on its OpenWorldEngineer coding benchmark with 67.2%
Mercor says it tested GPT-6.2 Astra and GPT-6.2 Sol before launch, and that Astra leads its new OpenWorldEngineer benchmark, which covers 26 long-horizon engineering and research projects, with a 67.2% mean score. It lists Claude Fable 5.2 at 54.6% and GPT-6 Astra at 44.2%.
Mercor says Astra is first in all four domains: software development 65.9%, engineering 83.7%, machine learning 62.9% and science 60.4%. The figures are Mercor's own and the post does not give GPT-6.2 Sol's score.
The post had 815 points on X by November 18.
Sources
Get news like this as a push.
AI Tracker alerts you when something changes for the tools you use.