The permaspike effect is far worse than you realise
retained text
The new GPT-6 class models are honestly so mediocre compared to the actual 5.6 bunch that got released just a couple of months ago. I've noticed the economic scandal going on with OpenAI, and they are trying so hard to save their computing resources, while silently dumbing older models down to force users to make the switch over to modern, sterile versions of already-good models that have originally been trusted and loved by the majority of developers. Scam Altman seems to be diagnosed with Jungle's Bounty Syndrome, a type of symptom that I refer to as being self-proclaimed and egotistical, not caring about everyone and the environment around but the billionaire himself. If only he wasn't desperate to be so greedy and take away the LLM upgrades and flipped it to downgrades, then the GPT-6 series would be incredibly versatile, flexible and resourceful as you could plan your workflow around the sun, the moon and the galaxy together to make it as perfect as you could through vibe coding. Overall, I'm honestly a bit disappointed with the new sun (GPT-6 Sol) and moon (GPT-6 Luna) models so far. They are mostly considered 'sidegrades' because of their cost-efficiency but overall weaker cognitive and planning skills all around. And that's why I consider the "permaspike effect" phenomenon where strong LLMs gets intentionally toned down to make room for resources, such as a new model testing (assuming OpenAI is testing a budget version of GPT-6 Astra) or to be selfish and strip user freedom outright. I'm gonna switch to dedicated Chinese models like Deepseek & Kimi as they perform immensely and could honestly be better than the sloppy downgrades we were given. Kimi K2.7/K3 for usual coding tasks (depending on how complex the coding task is) while Deepseek 4.1 Flash plans around and fix minor mistakes Kimi could have let out.
document fetched 2026-09-30 00:29 · canonical https://reddit.com/r/OpenAI/comments/1wtlrth/the_permaspike_effect_is_far_worse_than_you