OpenAI announces GPT‑5.4, its most powerful model that excels at professional tasks

OpenAI has released GPT-5.4, a frontier model with improvements in reasoning, coding, and professional document creation.

Today, OpenAI launched GPT‑5.4, its most powerful and efficient frontier model. With this new model, OpenAI has brought the recent advances in reasoning, coding, and agentic workflows seen in the Codex series models to a general-purpose model that excels at professional tasks, including creating spreadsheets, presentations, and documents. As a result, OpenAI claims that GPT-5.4 can perform complex work more accurately, effectively, and efficiently than before.

This new model is available in ChatGPT, Codex, and via API in two variants. In ChatGPT, this model is available as GPT‑5.4 Thinking and GPT‑5.4 Pro. When users ask a question to this model, they will get an upfront plan of its thinking, allowing them to adjust the process even while it is working on the response. OpenAI also highlighted that GPT-5.4 will perform deep web research queries better than GPT-5.3 Instant.

GPT‑5.4 also comes with significantly improved computer-use capabilities, achieving 75% in the OSWorld-Verified Computer Use benchmark. Also, for the first time, OpenAI is delivering 1M tokens of context support in preview. In GDPval, GPT‑5.4 achieved a new state-of-the-art score of 83%. On MMMU-Pro, GPT‑5.4 achieved an 81.2% success rate without tool use, an improvement over Gemini 3.1 Pro's 80.5%.

One of the main complaints about OpenAI's GPT models was the ability to create visually better documents and presentations. OpenAI has tried to fix this problem in this model and claims that GPT-5.4 delivers stronger aesthetics, greater visual variety, and more effective use of image generation. Apart from the improved performance, OpenAI mentioned that GPT-5.4 is its most token-efficient reasoning model.

As you can notice in the benchmark comparison table below, the GPT-5.4 series performs better than competing models from Anthropic and Google.

Developers can also take advantage of the new /fast mode in Codex, which will deliver up to 1.5x faster token speeds with this new GPT‑5.4 model without losing intelligence. In the API, instead of taking all tool definitions in the prompt upfront, this new model receives a lightweight list of available tools along with a tool search capability.

GPT‑5.4 Thinking is already available for ChatGPT Plus, Team, and Pro users, replacing GPT‑5.2 Thinking. However, GPT‑5.2 Thinking will be available under the legacy models list for another 3 months. The GPT‑5.4 Pro model is available to ChatGPT Pro and Enterprise plan users.

In the API, gpt-5.4 is priced at $2.50 per million input tokens and $15 per million output tokens. gpt-5.4-pro is priced at $30 per million input tokens and $180 per million output tokens.