Overview

• A restored model picker gives paid users access to GPT-4o, o3, GPT-4.1 and three GPT-5 modes—Auto, Fast and Thinking—alongside a lightweight Thinking mini fallback.
• OpenAI has capped GPT-5 Thinking at 3,000 messages per week for paid users and now reroutes excess queries to the Thinking mini variant.
• The GPT-4.5 model remains exclusive to Pro subscribers due to its high GPU cost, reflecting tiered access based on computational demands.
• Sam Altman’s compute-prioritization plan promises to keep paying ChatGPT subscribers and existing API customers at the front of the queue, support roughly 30% more API growth and roughly double the company’s compute capacity.
• In response to feedback that GPT-5 feels “cold,” OpenAI is planning updates to give the model a warmer tone and greater per-user personality customization.