OpenAI Unveils GPT-5: A New Era of AI Intelligence and Capability
OpenAI Unveils GPT-5: A New Era of AI Intelligence and Capability
OpenAI has officially launched GPT-5, the latest iteration of its flagship language model, making it accessible to all ChatGPT users, including those on the free tier. This release marks a significant advancement in artificial intelligence, with OpenAI claiming GPT-5 is substantially smarter, faster, more useful, and more accurate than its predecessors, boasting a notably lower hallucination rate.
A Leap Forward in AI Understanding
Sam Altman, CEO of OpenAI, described the leap from GPT-4 to GPT-5 as akin to the iPhone's transition to a Retina display, emphasizing its enhanced clarity and capability. "GPT-5 is the first time that it really feels like talking to an expert in any topic, like a PhD-level expert," Altman stated. This sentiment underscores the model's enhanced ability to provide in-depth knowledge and nuanced responses across a wide array of subjects.
Introducing a Family of Models
The GPT-5 launch introduces several model variants designed to cater to different needs and cost considerations. Alongside the core GPT-5 model, OpenAI has released GPT-5-mini, a lightweight version suitable for everyday queries, and GPT-5-nano, an even faster and more economical option primarily available through the API. For free users of ChatGPT, access will include both GPT-5 and GPT-5-mini. Paying subscribers to the Plus tier will benefit from significantly higher usage limits for these models. The premium $200 per month Pro tier offers unlimited access to GPT-5, along with GPT-5-pro, a more powerful version optimized for complex tasks, and GPT-5-thinking, which allows for extended processing of queries to achieve more comprehensive and accurate answers. Pro users will also retain access to legacy models.
Unified System and Intelligent Routing
A key innovation in GPT-5 is its unified system architecture. The chat interface now intelligently routes user queries to the most appropriate model based on the conversation's complexity, the user's subscription tier, and explicit user intent. This dynamic routing eliminates the need for users to manually select different models, streamlining the user experience. The router is continuously trained on real-world signals, including user feedback and performance metrics, to optimize model selection over time. Once a user reaches their allocated usage limits for the primary models, their queries are seamlessly transitioned to the GPT-5-mini model.
Enhanced Capabilities and Performance
GPT-5 demonstrates considerable improvements across various benchmarks. It boasts a 256,000-token context window, a significant increase from the 200,000 tokens in previous models, enabling it to better understand and retain context in long conversations, documents, or code. In coding assessments, GPT-5 reportedly outperforms its predecessors on benchmarks such as SWE-Bench Verified, SWE-Lancer, and Aider Polyglot. OpenAI showcased its capabilities by having GPT-5 generate a fully functional, interactive web application from a simple text prompt in a matter of minutes. Michelle Pokrass, a post-training lead at OpenAI, highlighted GPT-5's proficiency as a "great coding collaborator" and its effectiveness in agentic tasks, executing long chains of tool calls and following detailed instructions.
Reduced Hallucinations and Improved Safety
Addressing a persistent challenge in AI, OpenAI claims GPT-5 significantly reduces hallucinations—instances where the model generates false or inaccurate information. Alex Beutel, OpenAI's safety research lead, stated that the rates of deception have been "significantly decreased." System card evaluations indicate that GPT-5 models exhibit a 26 percent lower hallucination rate compared to GPT-4o when tested without web browsing access, and GPT-5-thinking shows a 65 percent reduction compared to o3. For prompts with potentially dual-use applications, GPT-5 employs "safe completions," aiming to provide helpful answers within safety constraints. OpenAI conducted over 5,000 hours of red teaming and collaborated with external organizations to ensure the system's robustness.
Developer and User Experience Enhancements
For developers utilizing the API, GPT-5 is priced at $1.25 per million input tokens and $10 per million output tokens. GPT-5-mini is available at $0.25/1M input and $2/1M output tokens, while GPT-5-nano is priced at $0.05/1M input and $0.40/1M output tokens, making it a competitive option against other cost-effective models. Starting soon, Pro users will be able to connect their Gmail, Google Contacts, and Google Calendar to ChatGPT, with other tiers gaining access at a later date. The company is also introducing customization features, including four new preset personalities—Cynic, Robot, Listener, and Nerd—that users can opt into to tailor ChatGPT's interaction style. These personalities are designed to reduce sycophancy and enhance user satisfaction.
Accessibility and Future Outlook
OpenAI reports a substantial user base, with nearly 700 million weekly active users of ChatGPT, 5 million paying business users, and 4 million developers using the API. Nick Turley, head of ChatGPT, expressed optimism that users, particularly those less familiar with AI models, will find GPT-5
AI Summary
OpenAI has launched GPT-5, the latest iteration of its flagship language model, making it accessible to all ChatGPT users, including free users. The company claims GPT-5 is smarter, faster, more accurate, and has a lower hallucination rate than previous models. CEO Sam Altman likened the advancement to the iPhone's Retina display, suggesting GPT-5 feels like conversing with a PhD-level expert. The launch includes several model variants: GPT-5-mini (lightweight), GPT-5-nano (API-only, faster and cheaper), and GPT-5-pro (for Pro subscribers, offering extended reasoning). Free users get GPT-5 and GPT-5-mini, while Plus subscribers receive higher usage limits. The $200/month Pro tier offers unlimited GPT-5 access, GPT-5-pro, and GPT-5-thinking for longer query processing. A key feature is the unified system where a router automatically selects the appropriate model based on query complexity and subscription tier, eliminating the need for users to manually choose. API pricing for GPT-5 is $1.25/1M input and $10/1M output tokens, with GPT-5-nano being cheaper than Gemini 1.5 Flash. Pro users will gain access to connect their Google services (Gmail, Contacts, Calendar) starting next week. GPT-5 boasts a 256,000-token context window, improving its ability to understand long conversations and documents. It also shows significant improvements in coding benchmarks, outperforming previous models. OpenAI demonstrated GPT-5's capability by having it generate a complex, interactive web application from a simple prompt. The model is also noted for being a strong coding collaborator and excelling at agentic tasks, executing long chains of tool calls effectively and following detailed instructions. Hallucination rates have been significantly decreased, with GPT-5 showing a 26% lower rate than GPT-4o and GPT-5-thinking having a 65% reduced rate compared to o3. OpenAI has implemented "safe completions" for dual-use prompts and conducted extensive red teaming. The company reports nearly 700 million weekly active ChatGPT users. GPT-5 is designed to be less sycophantic, use fewer emojis, and feel more like chatting with a helpful friend. Customization options include four new preset personalities: Cynic, Robot, Listener, and Nerd. GPT-5 Pro offers extended reasoning for the most challenging tasks, outperforming GPT-5 Thinking in expert evaluations. GPT-5 is now the default model in ChatGPT, replacing older versions, though paid users can still select legacy models. Free users will transition to GPT-5-mini upon reaching usage limits. The model was trained on Microsoft Azure AI supercomputers.