I’ve been noticing some weird behavior lately when using the o3 model through ChatGPT. It seems like OpenAI might be quietly switching o3 requests over to what could be GPT-5. There’s this mysterious model called “zenith” that showed up on LMArena, and people think it might be the real deal.
The thing that caught my attention is how the responses have changed. I tried giving it some really tough math problems that the original o3 model completely failed at before (like 0% success rate), but now it’s actually solving them correctly. The writing style is also totally different from what o3 used to produce.
Has anyone else noticed this change? It’s pretty sneaky if they’re doing this without telling users. Makes me wonder what other model switches they might be doing behind the scenes.
honestly, this sounds like typical openai behavior - they love stealth updates without announcements. i’ve been using o3 for coding and haven’t noticed major changes, but math isn’t my use case. that zenith thing is interesting though - could be internal testing that leaked? either way, we’re basically unknowing beta testers lol
Haven’t seen those specific changes with o3, but OpenAI’s pulled this before - silent updates without telling anyone. That “zenith” model on LMArena is weird though. Doesn’t sound like their usual naming at all. You mentioned better math reasoning? That’d be huge if it’s real. But here’s the thing - they’re constantly tweaking models based on how people use them. What you’re seeing might just be regular optimizations, not a whole new model. The sketchy part is how they don’t tell us what’s happening. If they’re switching which model handles our requests, we should know. Especially if you’re using this for work or research where you need consistency.