“GPT-5 is the best model we’ve built yet.” said OpenAI CEO Sam Altman during the company’s launch presentation for GPT-5. “[It is] great for a lot of things, but we think it’s going to be an especially important moment for businesses and developers”
OpenAI’s GPT-5 introduces a number of new features designed to improve usability, response speed, and personalization for both consumer and enterprise users. Released today, GPT-5 is now the default model in ChatGPT and is also available in the OpenAI API. Key upgrades include enhanced memory features, faster response times, customizable user settings, and improved coding and voice capabilities.
Memory and Personalization
GPT-5 supports persistent memory, allowing it to retain context across sessions and incorporate personal data where authorized. Starting this month, users can connect their Gmail and Google Calendar accounts to allow ChatGPT to provide schedule-aware responses and reference email content when planning or assisting with tasks. Memory-based personalization is available to Pro and Team tier users initially.
ChatGPT now also allows users to choose among different personality presets, including supportive, concise, and more casual tones. Paid users can further customize the interface with color themes and output verbosity.
Speedier Performance
According to OpenAI, GPT-5 delivers faster response times compared to GPT-4.1, particularly with its “minimal reasoning” mode, which prioritizes speed over depth. The model includes adjustable “reasoning_effort” settings, enabling users to control how deeply the model processes a prompt. Internal benchmarks show approximately 25–30% speed improvements over the previous generation.
The model’s standard context window has been expanded to 256,000 tokens in ChatGPT, enabling it to maintain consistency and reference across much longer interactions or documents. This extended memory enhances performance in summarization, legal, academic, and technical tasks.
Coding and Productivity Tools
GPT-5 includes upgrades to its coding abilities and is integrated into development tools like Cursor. According to Cursor’s team, GPT-5 is capable of identifying architecture-level issues, running end-to-end code generation, and iteratively debugging software within large codebases. Benchmarks show GPT-5 achieving 74.9% on SWE-Bench and 88.0% on AIDER Polyglot.
Developers using the OpenAI API can also leverage GPT-5’s structured output, multi-step tool use, and preambles for better integration in software workflows. The model’s performance has led to adoption in IDEs and custom AI developer assistants.
Voice, Visuals, and Multimodal Enhancements
GPT-5 expands support for voice interaction with improved natural language output and new adjustable styles. Users can specify answer formats such as “one-word” or “brief explanation,” and select tone and pace during speech. A new study mode allows for guided learning, including language practice with dynamic feedback.
Multimodal capabilities have also improved. GPT-5 supports advanced image understanding, visualization generation, and cross-modal reasoning. It scored 84.2% on the MMMU benchmark, which includes visual comprehension tasks.
Use Cases in Health and Education
The model demonstrates enhanced capabilities in health and education contexts. For instance, users can input medical documents or exam questions and receive more thorough, context-aware explanations. OpenAI reports that GPT-5 scored higher than prior models on its internal health benchmark and was developed with support for improved factual reliability and user empowerment.
Availability and Access
GPT-5 is now the default model in ChatGPT for Plus, Pro, and Team users. Free-tier users begin with GPT-5 and are transitioned to GPT-5 Mini after usage thresholds are met. API pricing varies by model: $1.25 per million input tokens for the full GPT-5 model, $0.25 for GPT-5 Mini, and $0.05 for GPT-5 Nano.
ChatGPT Enterprise and EDU customers are being rolled out GPT-5 in phases, with broad availability expected by the end of the month. The model is also integrated into Microsoft Copilot tools and other OpenAI-supported applications.
This launch coincides with OpenAI’s release of two open-weight language models, gpt‑oss‑120B and gpt‑oss‑20B, the company’s first freely downloadable models since GPT‑2.
