greg confirms 5.5 is spud, a new pre-trained model two years in the oven.
whilst it’s already powerful, we’re still early in the pipeline, so expect more impressive leaps within a month.
as more post-training and rl is done on spud.
in three months, expect earth-shattering advances in what’s possible.
this is indeed a new paradigm.
🚨 GPT-5.2 is finally OUT!
I’ve got early access and it’s a significant upgrade over 5.1 especially in visual understanding and reasoning.
The benchmark jumps vs 5.1 speak for themselves:
SWE-Bench Pro: 50.8% → 55.6%
GPQA Diamond: 88.1% → 92.4%
AIME 2025: 94.0% → 100%
ARC-AGI-2: 17.6% → 52.9%
It’s also surprisingly good with 3D/physics, so of course I had to throw a harder version of our favorite vibe test at it