GPT-4o Returns and GPT-5 Gets New Modes After User Backlash

Just one week after GPT-5’s high-profile launch, OpenAI has made major changes in response to widespread user feedback. Complaints about inconsistent performance, missing older models, and a confusing auto-switching system pushed the company to adjust — bringing GPT-4o back for paying users and giving GPT-5 new, user-controlled modes.

Why the Change Happened

When GPT-5 debuted, it featured an “auto-switcher” designed to choose between fast answers and deeper reasoning without user input. In practice, many found it unreliable — sometimes switching when it shouldn’t, slowing responses, or producing less accurate results.

The removal of GPT-4o and other familiar models also frustrated users, especially those who preferred their tone or consistency. Feedback was swift, and OpenAI acted quickly to address it.

What’s New for GPT-5 Users

OpenAI has rolled out two key updates:

1. GPT-4o Restored for Paying Users

   • Available by default for Plus, Pro, Team, Enterprise, and Edu accounts.

   • No more enabling “legacy models” to access it.

2. Three GPT-5 Modes You Can Control

   • Auto – GPT-5 chooses between speed and deep reasoning.

   • Fast – Prioritizes speed for quick, low-risk jobs like summaries, unit conversions, and simple calculations.

   • Thinking – Designed for complex, high-stakes work such as multi-step engineering calculations, advanced code reviews, and detailed safety audits. This mode uses a 196,000-token context window.
     

Limits on Thinking Mode

Paying users can send 3,000 messages per week in GPT-5 Thinking mode. After that, usage switches to GPT-5 Thinking Mini, which is faster but less powerful for long reasoning tasks.

These changes bring flexibility and reliability back to ChatGPT workflows. Engineers can:

• Use Fast mode for rapid drafting, formatting, and quick checks.

• Switch to Thinking mode for high-precision calculations, multi-step design problems, or detailed code reviews.

• Fall back on GPT-4o if GPT-5’s style or performance isn’t ideal for a specific task.
  

OpenAI’s quick response to GPT-5’s rocky debut shows how user feedback shapes product direction. For engineers, the result is more control over performance, better handling of complex work, and the return of a trusted model in GPT-4o.

For more AI insights like this, subscribe to the Singularity Newsletter → https://www.singularityengineering.ca/general-4