VulcanBench author calls GPT-6 Astra results unlike anything seen before
A benchmark author says early GPT-6 Astra results from VulcanBench are very special and unlike any model behavior seen before, and that he will study the reasoning traces before releasing full results.
I am starting to see initial results come in from my GPT-6 Astra benchmark with @VulcanBench. Don’t want to give much away yet, but I can tell you, this is something very special. An honor to benchmark at the frontier, and a greater challenge than anything I have ever done in my life. Congrats to the whole team at OpenAI, some of the results, and behaviors I’m seeing with this model are like nothing I have ever seen before. Might be delayed getting all the results out as I’ll be studying the traces, feels like the closest I’ve been to watching Einstein think. And I want to really understand how it’s doing what it does. Mind blown.