xiyu|Sep 04, 2026 00:40
OpenAI releases the next-generation flagship model GPT-6 Astra, along with the system card, and has started rolling it out; it achieved a 99.9% score on the ARC-AGI-3 benchmark.
GPT-6 Astra's evaluation on ARC-AGI-3 used the Responses API workflow, where GPT-5.6 Sol's estimated score under the same workflow is around 30%, making it hard to directly compare with the publicly noted 7.8%. This marks OpenAI's first full-version flagship upgrade since GPT-5.
Share To
Timeline
HotFlash
APP
X
Telegram
CopyLink