ChatGPT 5: What’s New and How It Works

OpenAI launched GPT-5 on August 7, 2025, and rolled it into ChatGPT with smarter reasoning and a new routing system that chooses the right mode for your task. If you’ve seen the note “Used GPT-5” under some replies, that’s ChatGPT automatically picking the GPT-5 family when it helps most.

Key Changes at a Glance

• Two modes, one picker: ChatGPT can run Chat or Thinking for GPT-5, and Auto decides when to think longer for complex work.
• Better reasoning & memory: GPT-5 improves multi-step reasoning and long-context handling, especially in projects and multi-turn workflows.
• Responses API: Devs get stateful-by-default conversations, hosted tools, and easier agentic flows—this is how apps tap GPT-5’s strengths.
• Capacity updates: Plus/Enterprise plans gained higher weekly limits and options to force Thinking or keep it fast.

How the Modes Work

• Auto: Let ChatGPT choose between speed and depth. It switches to GPT-5 Thinking when your request needs more deliberate reasoning.
• Thinking: You can explicitly select deeper reasoning for analysis, planning, coding, or debugging.
• Fast/Instant (where available): Prioritizes quick replies using lighter GPT-5 variants.

You may sometimes see a banner like “Used GPT-5”. That’s ChatGPT’s router picking GPT-5 (or GPT-5 Thinking) for the best result in that turn.

What Developers Get

If you build with OpenAI, the Responses API brings persistent tool state, cleaner multi-modal calls, and smoother agent workflows—designed with GPT-5 in mind. This reduces boilerplate and keeps conversation/tool context around without extra plumbing.

Why It Matters

• Users: Smarter default answers with fewer toggles to manage; you can still force modes for control.
• Teams: Enterprise tiers offer higher message capacity and admin controls while retaining GPT-5 quality.
• Builders: The API shift means faster iteration on assistants that reason, search, code, and use tools.

Quick Tips

• For research or complex planning, switch to Thinking if Auto isn’t deep enough.
• For speed or casual chat, stick to Fast/Instant (when offered in your plan).
• In apps, adopt Responses API to unlock GPT-5’s stateful workflows with less code.