OpenAI on Thursday released its latest family of frontier artificial intelligence (AI) models. Named GPT-5, it is said to offer a significant improvement in performance across coding, health, math, and visual perception. The San Francisco-based AI firm said GPT-5 features significant improvements in coding, mathematics, writing, health, visual perception, and more. Unlike previous model releases, this flagship model is being released to users across all subscription tiers, as well as those on the free tier.
GPT-5 Plans, Availability
In a blog post, OpenAI said that GPT-5 is currently rolling out to all users, including ChatGPT Plus, Pro, and Team subscribers, and those on the free tier. Enterprise and Edu users will get access to the model next week. Paid subscribers also have access to the AI model in the Codex CLI.
OpenAI confirmed that GPT-5 will replace GPT‑4o, OpenAI o3, OpenAI o4-mini, GPT‑4.1, and GPT‑4.5 for logged-in users. This will eliminate the need to choose different models for different tasks from the model picker. The company says the new AI model can intuitively understand when to use reasoning and when to generate quick responses. However, paid subscribers will have access to a separate GPT-5 Thinking, which is a deeper reasoning model for harder problems.
ChatGPT Plus users get extended access to the AI model, whereas the ChatGPT Pro plan, priced at $200 (roughly 17,500), offers unlimited access, along with availability of GPT-5 Pro as well. Meanwhile, the company will begin its rollout with the ChatGPT Enterprise and Edu plans on August 14.
As per the OpenAI, the use of GPT-5’s API will cost $1.25 (roughly Rs.110) per one million to input tokens and $10 (roughly Rs. 880) per one million to output tokens. GPT-5 Nano is the cheapest among the bunch, costing $0.05 (roughly Rs. 4) per one million to input and $0.40 (roughly Rs. 35) to output tokens.
GPT-5 Features
The San Francisco-based company said that GPT-5 is a unified system of LLMs which offer improved performance across coding, health, math, visual perception, and writing. It is claimed to be less prone to hallucinations and pretending to know things, with GPT-5 Thinking especially proficient in this. Instead, the AI system can decide between when to respond quickly and think longer to deliver accurate and comprehensive responses.
OpenAI claimed that GPT-5 outperforms all of its previous models in academic and human-evaluated benchmarks. Highlighting its coding proficiency, it is said to have scored 74.9 percent on the SWE-bench Verified. This score is higher than GPT-4.1 which registered a 54.6 percent score, positioning GPT-5 as the company’s most advanced model for coding purposes.
Coming to the improvements in this generation, GPT-5 gets a real-time router that switches between everyday and reasoning modes. The company said this router is continuously trained on real signals such as when users switch models, preference rate for responses, and others to make the experience more intuitive. OpenAI also highlighted that hallucinations have been reduced in the LLM while instruction adherance has been increased. The model is also said to show improved performance in writing, coding, and health.
In coding, the model is said to offer improved front-end generation and debugging larger repositories. It can now create responsive websites, apps, and games with a single prompt, the company claimed. When it comes to writing, GPT-5 is said to better handle structural ambiguity, making it balance form and expressiveness.
Health is a particularly important improvement area for the LLM. OpenAI said, “Compared to previous models, it acts more like an active thought partner, proactively flagging potential concerns and asking questions to give more helpful answers.” Additionally, it is said to better adapt to the user’s context, knowledge level, and geography to provide tailored responses.
Apart from this, the company is also launching a research preview of four new preset personalities in GPT-5-powered ChatGPT. These four options, Cynic, Robot, Listener, and Nerd, are opt-in experiences that match the user’s communication style, and are adjustable. These are currently available for text chat, but later will be added to Voice Mode.
On safety, OpenAI said it evaluated the model for 5,000 hours, and on the basis of the feedback from red teamers and internal staff, it has implemented strong safeguards to “sufficiently minimise the associated risks.” To handle potentially unsafe queries, OpenAI has trained GPT-5 on a new form of safety training called safe competitions, which replaces the refusal-based safety training used before. It basically educates the LLM to provide a helpful answer up to the point it is possible to stay within its outlined safety boundaries.
If it detects a need to refuse, it will provide a transparent response along with the reason, and provide safer alternatives. The company said that this approach is more nuanced and mitigates unnecessary overrefusals.
Microsoft Incorporates GPT-5
Microsoft has announced the incorporation of GPT-5 in its Copilot, 365 Copilot, enterprise, and conusmer offerings. It improves upon Microsoft 365 Copilot’s reasoning capabilities, enabling it to hold longer conversations and better understand the context. Further, Copilot is now better equipped to handle complicated problems, as per the company.
Customers with Microsoft 365 Copilot licenses can reason over documents, emails, and files. It is also being made available in Copilot Studio for building custom agents. Users can select GPT-5 in custom prompts.
Alongside, GitHub is onboarding GPT-5 in all paid GitHub Copilot plans. It is claimed to enable developers to write, text, and deploy code across GitHub Copilot Chat on github.com, Visual Studio Code and GitHub Mobile through the model picker. Meanwhile, developers in VS Code can also use it to develop agents without exiting the code editor.
Lastly, developers using Azure AI Foundry also get access to OpenAI’s latest family of AI model and take advantage of the router which leverages the AI-powered orchestration layer to determine the optimal model to use after evaluating each prompt for complexity, performance needs and cost efficiency of the task.
