OpenAI has launched GPT-5 and rolled it out to all ChatGPT users—including the free tier—while also opening up an API for developers on the same day. The company claims the new model is faster, more accurate, and delivers deeper reasoning, with hallucinations reduced by up to 45% compared to GPT-4o.
Key launch facts
- Date & event: GPT-5 debuted at OpenAI’s “Summer Update” press briefing on August 7, 2025.
- CEO remarks: Sam Altman called GPT-5 “the best model in the world” and “a significant step toward AGI.”
- Global focus: Altman highlighted India as OpenAI’s second-largest and fastest-growing market.
What’s new under the hood
- Advanced reasoning: GPT-5 uses an internal chain-of-thought router that decides on the fly whether to answer quickly or “think” longer for tough tasks.
- Coding power: It can generate full software applications on demand and outperforms earlier models on coding benchmarks.
- Safety upgrades: With 5,000 hours of evaluation and a “safe completion” system, OpenAI says risky outputs are reduced; hallucinations are 45% lower than GPT-4o and 80% lower than the O3 series during reasoning.
- PhD-level expertise: OpenAI says responses now reach expert-level depth across many domains.
Model family & pricing
- gpt-5: Full reasoning model and the default inside ChatGPT.
- gpt-5-mini: Built for cost-sensitive apps; kicks in automatically when free-tier limits are reached.
- gpt-5-nano: Optimized for ultra-low latency and the fastest responses.
- gpt-5-chat / GPT-5 Pro: For Enterprise and Pro users with extended reasoning and higher limits; Pro access priced at $200 per month.
API launch details
- Availability: API released the same day in three sizes—gpt-5, mini, and nano.
- Pricing: Base price set at $1.25 per million input tokens and $10 per million output tokens.
Access tiers inside ChatGPT
- Free: Limited GPT-5 usage before automatic fallback to gpt-5-mini.
- Plus ($20/month): Higher GPT-5 limits for regular users.
- Pro ($200/month): Unlimited GPT-5 access, plus GPT-5 Pro for advanced workloads.
- Team, Enterprise, Edu: GPT-5 becomes the default model next week.
Why it matters
- Momentum check: GPT-5 is seen as a key test of whether AI progress is still accelerating after GPT-4’s 2023 debut.
- From chatbot to agent: Signals a major shift toward agentic AI that can plan and complete multi-step tasks on its own.
- Big tech race heats up: Raises the stakes against Anthropic, Google DeepMind, and xAI in the push toward practical AGI.
Early reception
- Reviewers say going back to GPT-4 feels “miserable,” pointing to clear gains in speed and accuracy.
- Policy voices caution that capability is outpacing governance, urging stronger safeguards alongside rapid deployment.
Pingback: Elon Musk Claims His Grok 4 Beat GPT-5 Before It Even Launched – But Reddit Users Reveal The Real Story
Pingback: OpenAI's GPT-5 Flops Hard, Sparks User Fury – They Finally Canceled GPT-4 Shutdown on Fans' Desperate Requests!