OpenAI surprises users: GPT-5 now default - what changes?

OpenAI has begun rolling out GPT-5 as the default model in ChatGPT, promising higher factual accuracy and clearer reasoning for users on every plan. The company released the upgrade earlier today.

Rollout schedule

GPT-5 replaces GPT-4o effective immediately for Free, Plus, Pro, and Team customers. Enterprise and Education tenants follow next week. Free users receive full GPT-5 responses until they hit a daily usage limit, after which requests route to a lighter “GPT-5 mini” model.

Access tiers and limits

• Free: Full GPT-5 output until caps are reached; then GPT-5 mini.
• Plus: Higher caps than Free.
• Pro: Highest caps and “GPT-5 Pro” for compute-heavy tasks.
• Team: Same model availability as Pro, shared across workspaces.
• Enterprise & Education: Rollout begins next week.

Performance improvements

• 45% fewer factual errors than GPT-4o, rising to 80% with deeper reasoning enabled.
• Automatic adjustment of response speed or depth based on prompt complexity.
• Step-by-step reasoning available when users ask the model to “think hard.”
• Multimodal input now supports images and charts for analysis.
• Faster front-end code generation and automated debugging.
• Revised safety layer prefers partial answers over blanket refusals and reduces flattery.

Detailed evaluation results are published in OpenAI’s Evaluations post.

Background and timeline

Since GPT-3.5 launched in November 2022, OpenAI has iterated quickly: GPT-4 arrived in March 2023 and GPT-4o in May 2024, each adding speed, multimodal capability, and coding assistance. GPT-5 shifts the emphasis toward accuracy and broader availability. OpenAI calls the release a trust-building step ahead of more significant system changes expected later this year.

Source citations

• OpenAI blog post: “Introducing GPT-5”
• GPT-5 product page: released