Press Space to continue
GPT 5.2 achieves higher accuracy with more compute time
Press Space to continue
OpenAI just dropped that GPT 5.2 is #1 on the METR eval, finishing 6.5hr long human tasks half of the time, but.. It takes ~10x the working time per task of Gemini 3 Pro and ~25x of Opus 4.5. The models are smarter, but by using insane amounts of test-time compute reasoning!
open on xViews and clicks count once per visitor per day. Clicks are opens of this post on X.