On Thursday, OpenAI unveiled its latest artificial intelligence (AI) model family, titled GPT-5. This new model reportedly brings considerable advancements in various domains, including coding, health care, mathematics, and visual perception. The San Francisco-based company highlighted that GPT-5 incorporates major enhancements over its predecessors, now accessible to users across all subscription tiers, including those on the free plan.
GPT-5 Plans and Availability
According to a blog post from OpenAI, GPT-5 is being gradually rolled out to all users, encompassing ChatGPT Plus, Pro, and Team subscribers, as well as those utilizing the free tier. Users from enterprise and educational sectors will gain access to the model next week. Additionally, paid subscribers can utilize the new AI model via the Codex CLI.
The introduction of GPT-5 will phase out earlier models, including GPT‑4o, OpenAI o3, OpenAI o4-mini, GPT‑4.1, and GPT‑4.5 for logged-in users. This change streamlines the process by eliminating the necessity to select different models for various tasks. OpenAI noted that the new AI model is adept at discerning when to employ reasoning versus when to generate rapid responses. Moreover, paid subscribers will have access to a refined version called GPT-5 Thinking, designed for more complex problem-solving.
ChatGPT Plus subscribers will benefit from added access, while those enrolled in the ChatGPT Pro plan, which costs $200 (approximately 17,500), will have unlimited access to GPT-5, alongside additional features in GPT-5 Pro. The company plans to initiate its rollout for the ChatGPT Enterprise and Edu plans starting August 14.
The pricing for GPT-5’s API is set at $1.25 (around Rs.110) for each million input tokens and $10 (approximately Rs. 880) for every million output tokens. The most economical option, GPT-5 Nano, will cost $0.05 (about Rs. 4) for input tokens and $0.40 (around Rs. 35) for output tokens.
GPT-5 Features
OpenAI described GPT-5 as a cohesive system of large language models (LLMs) that deliver enhanced performance across various domains. It has been engineered to be less susceptible to generating inaccuracies and misrepresenting knowledge. Notably, the GPT-5 Thinking variant excels in maintaining accuracy through thoughtful responses.
The company asserts that GPT-5 surpasses its earlier models in both academic assessments and human evaluations. Particularly highlighting its coding capabilities, GPT-5 achieved a score of 74.9 percent on the SWE-bench Verified, significantly exceeding the 54.6 percent score of GPT-4.1, thus establishing GPT-5 as the most advanced coding model to date.
Among the notable improvements in GPT-5 is a real-time router that alternates between casual and reasoning modes. OpenAI explained that this router is continuously informed by real-time user feedback, such as model switching and response preference rates, aiming to create a more intuitive user experience. The company also noted a decrease in hallucinations within the model, along with enhanced adherence to user instructions.
In the coding domain, GPT-5 offers better capabilities for front-end development and debugging larger codebases. It is now reportedly able to create responsive websites, applications, and games using a single prompt. When it comes to writing, the model is designed to manage structural ambiguity more effectively to ensure a balance of form and expressivity.
Improvements in health-related capabilities are particularly emphasized. OpenAI stated, “Compared to previous models, it behaves more like an active thought partner, proactively identifying potential issues and engaging in dialogue to provide more helpful responses.” It is also designed to adjust according to the user’s context, expertise, and geographical location for better-tailored responses.
Additionally, the company is rolling out a research preview of four new preset personalities in the GPT-5-powered ChatGPT. The options—Cynic, Robot, Listener, and Nerd—are adjustable experiences that align with users’ communication styles. Currently, these personalities are available for text chat, with plans to expand into Voice Mode in the future.
Regarding safety, OpenAI conducted 5,000 hours of evaluation, incorporating insights from red team reviewers and internal feedback. The company implemented robust safeguards aimed at drastically reducing inherent risks. The new training methodology is dubbed safe competitions, which replaces former refusal-based safety training. This approach allows the model to provide helpful responses while remaining within its safety parameters.
In cases where a refusal is necessary, GPT-5 offers transparent reasoning along with safer alternatives, thereby addressing concerns about overrefusals.
Microsoft Incorporates GPT-5
Microsoft has announced integration of GPT-5 into its Copilot features within Microsoft 365, enhancing both enterprise and consumer offerings. This upgrade bolsters the reasoning capabilities of Microsoft 365 Copilot, enabling it to engage in lengthier conversations and better comprehend context. The improved Copilot is now said to handle more complex issues effectively, according to company statements.
Customers with Microsoft 365 Copilot licenses can now leverage GPT-5 to analyze documents, emails, and files. Furthermore, it will be accessible through Copilot Studio for creating custom agents, allowing users to select GPT-5 for personalized prompts.
In addition, GitHub is adopting GPT-5 across all paid GitHub Copilot plans, facilitating developers in writing, editing, and deploying code. The model will be available through GitHub Copilot Chat on github.com, Visual Studio Code, and GitHub Mobile, aiding developers within the code editing environment.
Finally, developers utilizing Azure AI Foundry will also have the opportunity to engage with OpenAI’s latest AI model family, benefiting from a router designed to optimize model selection based on task complexity, performance needs, and cost efficiency.
