Overview

• OpenAI has reverted its paid default model back to GPT-4o, a change announced by CEO Sam Altman on X and reflected in the release notes.
• Users can now switch GPT-5 between Auto, Fast and Thinking modes to balance speed, cost and depth of response.
• Thinking mode is limited to 3,000 messages per week and GPT-5’s context window is officially capped at 196,000 tokens, though some reports suggest higher limits.
• Beta connectors for services such as Google Docs, Notion, Canva and GitHub are now available outside Europe to link GPT-5 directly with user data.
• GPT-5’s updated safety guardrails evaluate potential outputs rather than just inputs but can still be bypassed via custom instructions, prompting further refinements to tone and security measures.