Overview

• GPT-4o returns as the default legacy model for all paying ChatGPT subscribers, with toggles added to restore GPT-4.1, o3 and o4-mini
• Three GPT-5 settings—Auto, Fast and Thinking—offer users explicit control, with Thinking capped at 3,000 messages per week and overflow shifting to a Thinking mini mode
• The GPT-5 Thinking mode now supports a 196,000-token context window, enabling much longer conversations and document handling
• OpenAI will prioritize compute for existing paying ChatGPT and API customers and intends to roughly double its infrastructure capacity over the next five months
• New third-party connectors for services such as Dropbox, Microsoft Teams, Box, Notion and HubSpot have been rolled out to paid tiers, excluding certain regions