GPT 6 (non Astra) is performing much worse than expected?
retained text
Using the GPT 6 class of models have been straight up frustrating. I've not been this frustrated with models since the 5.3 era. Sol is literally misinterpreting prompts, rewriting the UI (changing graphs into card based lists, etc), and generally building things as if there was no thought behind it at all (it still takes a while, so I don't think it's a thinking budget thing). It's not that it isn't able to complete the task. It just feels like a genie that's interpreting what I say in a super passive aggressive way. It's making mistakes which I've not seen in the frontier for \~6 months now. I can't even trust it with long running tasks anymore. Is this just me? My (incorrect) headcanon is that they renamed 6 Sol to Astra, Terra to Sol, and kept Luna as is. Explains why there's no Terra perfectly.
document fetched 2026-09-27 00:28 · canonical https://reddit.com/r/OpenAI/comments/1wquho6/gpt_6_non_astra_is_performing_much_worse_than