OpenAI introduced ChatGPT-5

OpenAI has unveiled GPT-5, the flagship AI model that will be the foundation for the next generation of ChatGPT. It is the company’s first “unified” model, combining the logical capabilities of the o-series with the responsiveness of the GPT models.

With the GPT-5, ChatGPT goes beyond just a conversational assistant. Now it can perform real-world tasks: building apps, managing your calendar, writing research briefs, and more.

And now it can perform real-world tasks like creating apps, managing your calendar, writing research briefs, and more.

The model automatically selects the best way to respond with a real-time router – no more manual configuration.

OpenAI CEO Sam Altman called GPT-5 “the best model in the world” and said it is a “significant step” toward creating AGI, an AI that outperforms humans in most economically relevant tasks.

“Something like GPT-5 would have been just unthinkable at any other time in history,” Altman said.

Since ChatGPT launched in 2022, it has become one of the most popular products in the world, reaching 700 million users per week. Now GPT-5 is becoming the standard model even for free users.

• Unified architecture: combines logical thinking and fast responses.
• Agent Capabilities: GPT-5 can perform complex tasks on behalf of the user.
• Ease of use: automatic selection of response mode.
• Improvements in “vibro-coding”: GPT-5 can create full-fledged applications by text request.

• SWE-bench Verified (coding): 74.9% – higher than Claude Opus 4.1 (74.5%) and Gemini 2.5 Pro (59.6%).
• GPQA Diamond (PhD level science questions): 89.4% – leads over Opus 4.1 (80.9%) and Grok 4 Heavy (88.9%).
• HealthBench (medical queries): 1.6% hallucinations – many times less than GPT-4o (12.9%) and o3 (15.8%).
• Hallucination rate (overall): GPT-5 with «thinking» is 4.8%, while GPT-4o and o3 are around 20%.
• Tau-bench (online action simulation)
  • Airfare search: 63.5% (just below o3 – 64.8%)
  • Online shopping: 81.1% (just below Opus 4.1 – 82.4%)

  

  GPT-5 is less likely to make mistakes, better at distinguishing between malicious and safe requests, more likely to reject malicious tasks, and less likely to reject harmless ones.

  • Adaptive communication styles: you can choose from four new «characters» – Cynic, Robot, Listener, Nerd.
  • Free access to GPT-5: for the first time, a model of this level is available without payment.
  • Subscriptions:
    • Plus ($20/month): increased limits.
    • Pro ($200/month): unlimited access to GPT-5 and GPT-5 Pro.
    • Team, Edu, Enterprise: GPT-5 will become standard from the new week.
    • API access to GPT-5 in three sizes: gpt-5gpt-5-minigpt-5-nano
    • Customizable verbosity of responses.
    • Prices:
      • $1.25 per million input tokens
      • $10 per million output tokens

      This week, OpenAI also introduced gpt-oss, an open source reasoning model that can be run locally. However, GPT-5 is already setting a new performance standard in a number of tasks.

      While GPT-5 is comparable to its competitors in a number of ways, its real advantage will only become clear in use — and how it will affect the industry as a whole remains a major question.