First impressions of GPT6-Sol on ExtraHigh.. Meh.. its a lot dumber and makes mistakes quite a lot. I have to constantly remind it that its doing things incorrectly. But it seems to be much more token effecient...
So... imo gpt6 sol is utter garbage compared to 5.6 sol. It cant remember things, and just gives up without fixing the problem. Or worse says it cant be fixed and doesnt even try.
From Paweł Huryn on X
```I finally tested GPT-6 Sol on a real work. 2 repos. 105 hidden bugs. Find and fixed what you can.
It looks like a huge degradation. The results:
- GPT-6 Astra (max): 45
- GPT-5.6 Sol (max): 43.5
- Opus 5.5 (max): 41.7
- Muse Spark 1.3 (max): 32.2
- GPT-6 Sol (max): 29.3
```