OpenAI co-founder and CEO Sam Altman is openly addressing significant issues with yesterday’s launch of GPT-5, the company's new, flagship large language model (LLM) — promoted as its most powerful and advanced iteration yet.

Responding to user inquiries in a Reddit AMA (Ask Me Anything) thread and in a post on X this afternoon, Altman acknowledged a series of problems that have disrupted the GPT-5 rollout, such as malfunctioning model switching, subpar performance, and user confusion — leading OpenAI to partially reverse some of its platform changes and restore user access to previous models like GPT-4o.

“It was a bit rougher than we anticipated,” Altman wrote in response to a question on Reddit about the major GPT-5 launch.

Regarding the incorrect model performance charts displayed during OpenAI’s GPT-5 livestream, Altman explained: “People were working late and were exhausted, which led to human error. A lot comes together for a livestream in the last hours.”

AI Scaling Hits Its Limits

Power caps, rising token costs, and inference delays are reshaping enterprise AI. Join our exclusive salon to discover how top teams are:

• Turning energy into a strategic advantage

• Architecting efficient inference for real throughput gains

• Unlocking competitive ROI with sustainable AI systems

Secure your spot to stay ahead: https://bit.ly/4mwGngO

While he mentioned that the accompanying blog post and system card were accurate, these errors further complicated a launch already under scrutiny from early users and developers.

GPT-5 rollout updates: *We are going to double GPT-5 rate limits for ChatGPT Plus users as we finish rollout. *We will let Plus users choose to continue to use 4o. We will watch usage as we think about how long to offer legacy models for. *GPT-5 will seem smarter starting…

— Sam Altman (@sama) August 8, 2025

Problems with new automatic model router

One significant source of the issues, according to Altman, is OpenAI’s new automatic “router” that assigns user prompts to one of four GPT-5 variants — regular, mini, nano, and pro — with an optional “thinking” mode for more complex reasoning tasks.

On X, Altman disclosed that a crucial component of that system — the autoswitcher — was “out of service for a portion of the day,” causing GPT-5 to seem “significantly less intelligent” than intended.

As a solution, OpenAI states it’s making adjustments to the model decision boundary and will provide more transparency about which model is responding to a specific query.

A UI update is also forthcoming to allow users to manually activate thinking mode.

Additionally, Altman confirmed that OpenAI will now permit ChatGPT Plus users to keep using GPT-4o — the previous default model — after numerous complaints about GPT-5’s inconsistent performance. He mentioned on Reddit that the company is “trying to collect more data on the tradeoffs” before deciding how long to keep the legacy models available.

they brought back the old models!!!! (it's in settings under "legacy models") pic.twitter.com/0MN85jmicG

— ben (@benhylak) August 8, 2025

Yet many users, including OpenAI beta testers like Wharton School of Business professor Ethan Mollick, expressed confusion and disappointment at OpenAI unilaterally upgrading their ChatGPT experiences to GPT-5 and initially removing access to the older models.

Real-world performance lags behind hype

OpenAI’s internal benchmarks may indicate that GPT-5 leads among LLMs, but real-world users are experiencing something different.

Since the launch, users have shared numerous examples of GPT-5 making basic errors in math, logic, and coding tasks.

Data scientist Colin Fraser posted screenshots of GPT-5 incorrectly solving whether 8.888 repeating equals 9 (it does not, obviously), while another user showed it incorrectly solving a simple algebra problem: 5.9 = x + 5.11.

Other users also reported difficulties getting accurate answers to math word problems or using GPT-5 to debug its own presentation charts.

Developer feedback hasn’t been much better, with users posting images of GPT performing worse at “one-shot” certain programming tasks — completing them well with a single-prompt — compared to rival AI lab Anthropic’s new model Claude Opus 4.1.

And security firm SPLX found GPT-5 still suffers from serious vulnerabilities to prompt injection and obfuscated logic attacks unless its safety layer is strengthened.

OpenAI in the spotlight

With 700 million weekly users on ChatGPT, OpenAI remains the largest entity in generative AI by audience.

However, this scale has brought its own challenges. Altman noted in his X post that API traffic doubled over 24 hours following the GPT-5 launch, contributing to platform instability.

In response, OpenAI says it will double rate limits for ChatGPT Plus users and continue to adjust infrastructure as it collects feedback.

But the initial missteps — combined with confusing UX changes and errors during a high-profile launch — have provided an opportunity for competitors to gain ground.

The pressure is on for OpenAI to demonstrate that GPT-5 isn’t merely an incremental update, but a genuine advancement. Based on the initial rollout, many users remain unconvinced — for now.