ChatGPT rolls out GPT-5 speed controls and brings back 4o

Shifting gears on GPT-5

OpenAI just rolled out a fresh batch of ChatGPT controls. You can now toggle GPT-5 between Auto, Fast, and Thinking modes. Sam Altman notes that while most people will stick with Auto, having manual control gives power users exactly what they need.

There are concrete caps now: GPT-5 Thinking gets 3,000 messages per week with a 196k context window, falling back to GPT-5 Thinking mini once you hit the ceiling. Altman cautioned that these limits could shift over time depending on overall usage.

Paid users also get GPT-4o back in the picker by default, with promises of plenty of notice if it ever gets axed. A new "Show additional models" toggle in web settings unhides options like o3, 4.1, and GPT-5 Thinking mini, while the GPU-heavy 4.5 remains locked to Pro tiers. Meanwhile, OpenAI is tweaking GPT-5’s vibe to feel warmer than it is today, but less annoying than 4o, with Altman admitting the real goal is letting everyone customize model personality per user.

Why it matters

Balancing compute bills with vocal user pushback is a tightrope walk, and this update feels like OpenAI listening.

For power users, a 196k token window and 3,000 weekly queries make GPT-5 Thinking a reliable workhorse for deep coding sessions or parsing massive files. Having a mini fallback prevents workflow whiplash.

We think reinstating 4o and admitting personality is a personal preference proves one size never fits all in AI. Trying to force a single conversational tone on millions of people backfires, so building per-user custom personalities looks like the only sane path forward.