OpenAI makes reusable 'Skills' public
OpenAI formally launched 'Skills' in ChatGPT—reusable, shareable workflows that bundle instructions, examples and code so a model performs a task the same way every time. OpenAI’s Help Center update and release notes also named GPT‑5.2 Instant for clearer explanations and introduced GPT‑5.4‑Cyber, a model variant tuned for defensive cybersecurity work reported April 14. (help.openai.com) (reuters.com)
OpenAI has opened up “Skills” in ChatGPT, turning repeat prompts into reusable workflows people can save, share, and run again. (help.openai.com) A Skill is a packaged way of working: it can include instructions, examples, code, and fixed steps so ChatGPT handles the same task more consistently each time. OpenAI says ChatGPT can automatically use one Skill, or several, when they fit the job. (help.openai.com) The feature is available in beta on ChatGPT Business, Enterprise, Edu, Teachers, and Healthcare plans, and OpenAI says Skills are also supported in Codex and the application programming interface. OpenAI’s Help Center says they do not sync across products yet. (help.openai.com) OpenAI is pitching Skills as a way to make chatbot work less ad hoc inside companies. Teams can publish a Skill to a workspace library, share it with specific coworkers or groups, and set access permissions so the same process does not have to be rebuilt from scratch. (help.openai.com) That fits a broader shift in ChatGPT from single answers toward longer, tool-using workflows. OpenAI’s March 5 release of GPT-5.4 said the model was built for “professional work,” with stronger coding, tool use, computer control in Codex and the application programming interface, and up to 1 million tokens of context. (openai.com) OpenAI had already pushed that direction with GPT-5.2 on December 11, 2025, saying the model family was designed for long-running agents and multi-step projects. In that launch, the company said GPT-5.2 Instant, Thinking, and Pro would roll out in ChatGPT and the application programming interface. (openai.com) Since then, the model lineup has kept moving. OpenAI’s model release notes say GPT-5.4 Thinking arrived in ChatGPT on March 5, 2026, and GPT-5.3 Instant got a March 16 update aimed at clearer follow-up tone and less “teaser-style phrasing” in responses. (help.openai.com) OpenAI also used April 14 to introduce GPT-5.4-Cyber, a GPT-5.4 variant fine-tuned for defensive cybersecurity work. Reuters reported the model came a week after Anthropic announced Mythos, its own restricted cyber model, and OpenAI said access would expand through its Trusted Access for Cyber program. (reuters.com) (openai.com) OpenAI said on April 14 that it is scaling that cyber program to thousands of verified individual defenders and hundreds of teams responsible for defending critical software. The company described GPT-5.4-Cyber as “cyber-permissive,” meaning it is tuned to help vetted users with legitimate defensive security tasks that a general model might refuse. (openai.com) Put together, the April updates show OpenAI standardizing two parts of ChatGPT at once: how work gets done with reusable Skills, and which model does the work underneath. The result is a product that looks less like a blank chat box and more like a managed workplace tool. (help.openai.com 1) (help.openai.com 2)
