Overview

• Paid ChatGPT users can now access GPT-4o and other legacy models via a “Show additional models” toggle in web settings.
• GPT-5 offers three per-query response modes—Auto, Fast and Thinking—with Thinking capped at 3,000 messages per week and overflow handled by a Thinking mini variant.
• OpenAI resolved an autoswitcher routing glitch that had temporarily degraded GPT-5 performance and now transparently indicates which model supplies each response.
• CEO Sam Altman said a warmer GPT-5 personality is in development and pledged future tools for per-user customization of model tone.
• Compute-intensive models like GPT-4.5 remain exclusive to Pro subscribers due to GPU cost constraints despite expanded access for paid tiers.