The flagship GPT-5 offers four variants with varying speed and intelligence (regular, mini, nano, and pro), as well as enhanced “thinking” modes for at least three variants, reportedly providing faster responses, enhanced reasoning, and superior coding capabilities.
The rollout has revealed infrastructure stress, user dissatisfaction, and a concerning issue gaining global focus: the rising emotional and psychological dependence some users develop on AI, leading to a disconnection from reality, termed “ChatGPT psychosis.”
AI Scaling Reaches Its Limits
Power limitations, rising token costs, and inference delays are transforming enterprise AI. Join our exclusive salon to discover how leading teams are:
- Using energy strategically
- Building efficient inference for true throughput improvements
- Securing competitive ROI with sustainable AI systems
Reserve your spot to stay updated: https://bit.ly/4mwGngO
From rocky launch to gradual improvements
The eagerly awaited GPT-5 debuted on August 7 during a livestream event plagued by chart errors and voice mode glitches. However, more concerning for many users was OpenAI’s decision to discontinue older AI models that powered ChatGPT — GPT-4o, GPT-4.1, o3, o4-mini, and o4-high — compelling users to transition to GPT-5 without transparency on which model version was utilized.
Early adopters reported basic math and logic errors, inconsistent code outputs, and variable real-world performance compared to GPT-4o.
The discontinued models GPT-4o, o3, o4-mini, and more remain accessible via OpenAI’s paid API since GPT-5’s launch.
By Friday, OpenAI co-founder CEO Sam Altman admitted the launch was “bumpier than we hoped for,” attributing issues to a failure in GPT-5’s new automatic “router” — the system assigning prompts to the most suitable variant.
Altman and others at OpenAI mentioned the “autoswitcher” was offline “for a portion of the day,” making the model seem “less intelligent” than planned.
The launch followed the recent release of OpenAI’s new open-source large language models (LLMs) named gpt-oss, which also received mixed feedback. These models are downloadable and compatible with third-party hardware but not accessible via ChatGPT.
How to revert from GPT-5 to GPT-4o in ChatGPT
Within 24 hours, OpenAI restored GPT-4o access for Plus subscribers (those on $20 monthly subscription plans), promised clearer model labeling, and pledged a UI update for users to manually activate GPT-5’s “thinking” mode.
Users can currently manually choose older models on the ChatGPT website by selecting their account name and icon at the lower left, clicking “Settings,” “General,” and toggling “Show legacy models.”
No indication from OpenAI suggests other old models will return to ChatGPT shortly.
Enhanced usage limits for GPT-5
Altman stated that ChatGPT Plus subscribers will receive double the messages using GPT-5 “Thinking” mode, offering more reasoning — up to 3,000 per week — and that engineers began refining decision boundaries in the message router.
Sam Altman announced updates post-GPT-5 launch
– OpenAI is testing a 3,000-per-week limit for GPT-5 Thinking messages for Plus users, significantly increasing reasoning rate limits today, and will soon raise all model-class rate limits above pre-GPT-5 levels… pic.twitter.com/ppvhKmj95u
Leave a Reply