Usman Salis
⬤ OpenAI's GPT-5 lineup is showing some interesting patterns in the latest LMArena rankings. GPT-5.2 just hit its best score ever in the no-reasoning category, landing at 14th place. That's a solid jump when you compare it to GPT-5.1 at 18th and the original GPT-5 sitting at 26th. On the reasoning side of things, GPT-5.1-high is still running the show at rank 8, while GPT-5-high comes in at 19 and GPT-5.2-high sits at 17 in the style-controlled tests.
⬤ The benchmark data reveals two pretty different stories depending on which version you're looking at. The no-reasoning models have been climbing steadily, with GPT-5.2 marking the strongest performance in the current rankings. But the high-reasoning variants? That's where things get less predictable. GPT-5.1-high keeps its crown as the best in that group, while GPT-5.2-high hasn't quite matched that momentum, even though it's apparently doing well with math and coding tasks. All this comes from LMArena's December 2025 style control rankings.
⬤ These numbers matter because they're landing after OpenAI went on a bit of a release spree—launching GPT-5 back in August, then quickly rolling out GPT-5.1 and GPT-5.2 variants. The no-reasoning versions show consistent improvement across the board, but the high-reasoning models are bouncing around more depending on the specific tasks being tested. It's the kind of variation that happens when training methods keep evolving, and it gives a pretty clear picture of where each model actually shines.
⬤ What this really tells us is that GPT-5.2 is getting better at handling standard prompts, while GPT-5.1-high remains the go-to choice when you need serious reasoning power. The comparison highlights how fast OpenAI is iterating on the GPT-5 family, and it also shows that as these models get trained in different ways, performance can shift quite a bit depending on what you're asking them to do.
Usman Salis
