OpenAI has released GPT-5.4, a new frontier model built for long, complex tasks, native computer use, and higher-accuracy work. The model is rolling out across the OpenAI API, ChatGPT, Codex developer tools, and partner platforms this week.
OpenAI has addressed GPT-5.4 as their most capable model for professional jobs like spreadsheet modelling, creating documents and presentations, legal drafting, and multi-step automation.
The new model is designed to move models from short answers to real task completion. Its computer-use abilities let agents finish work across apps rather than just plan steps. Such capabilities change how companies can automate workflows and embed AI into day-to-day tools while raising new questions about safety, oversight, and cost control.
Benchmarks Achieved:
OpenAI’s GPT-5.4 has shown remarkable performance across industry-leading benchmarks for AI models. For knowledge work, it has scored 83.0% in GDPval. While outperforming coding, the model attained 57.7% in SWE-Bench Pro.
In the leading benchmark for computer use and vision, OSWorld-Verified, the model scored 75.0%. It has accordingly achieved 54.6% in Toolathlon and 82.7% in BrowseComp, outperforming previous GPT versions.
Key Features to Watch in GPT-5.4:
- GPT-5.4 includes very large context windows for document-heavy work and long conversations.
- It enables native computer usage so that agentic flows can operate across files, tools, and web pages.
- The model also has improved coding and multi-step task execution, powering better Copilot experiences.
- A Pro variant for higher precision and stricter performance limits for enterprises.
Pricing of GPT-5.4
OpenAI prices GPT-5.4 on a per-token basis. Standard rates start at $2.50 per 1 million input tokens, $0.25 per 1 million cached input tokens, and $15 per 1 million output tokens. Larger context sessions and the Pro tier carry higher rates and different rate limits.
How to Access?
Developers can access the model via the OpenAI API and Codex. The model is also appearing inside ChatGPT as GPT-5.4 Thinking and, for higher tiers, GPT-5.4 Pro.
So, you can access the model by specifying it in the API or by selecting it in ChatGPT’s model picker, if your plan includes it.
Additionally, you can try the latest model in the OpenAI sandbox, enable it in ChatGPT if your subscription permits, or pick it in GitHub Copilot if you are on a paid Copilot plan. Enterprises wanting production support can explore availability through Microsoft Foundry and Azure’s model catalog.
GPT-5.4 pushes agentic, document-heavy AI toward real work. It is powerful but costly, so teams should test with clear guardrails and a plan for monitoring and budget control.
Stay current with the latest tech updates and trends with our buzzing news headlines!
Recommended For You:
How to Use the Indeed App in ChatGPT for Your Job Search
Examining Gemini Nano AI and Its Features: Is It Better than ChatGPT?