© ROOT-NATION.com - Use of content is permitted with a backlink.
OpenAI has introduced GPT-5.4, its most powerful model for solving professional tasks. The new development combines the latest advances in logical thinking, programming, and agent-based workflows previously implemented in the Codex series models and embodies them in a general-purpose model. Particular attention is paid to efficiency when working with business tasks, such as creating spreadsheets, presentations, and documents. According to OpenAI, GPT-5.4 is capable of performing complex operations more accurately, efficiently, and with less resource consumption than previous versions.
The model is available in three environments: ChatGPT, Codex, and via API in two versions. In ChatGPT, it is presented as GPT-5.4 Thinking and GPT-5.4 Pro. When processing a query, the user receives a preliminary plan of the model’s reasoning, which allows making adjustments directly in the process of forming an answer. In addition, OpenAI noted that GPT-5.4 outperforms GPT-5.3 Instant in performing deep web search queries.
Among the technical improvements, it is worth highlighting the significantly improved computer usage capabilities: the model reached a score of 75% in the OSWorld-Verified Computer Use benchmark. For the first time in the OpenAI lineup, support for a context window with a volume of 1 million tokens in the preview mode was introduced. In GDPval, the model set a new record with a score of 83%. In the MMMU-Pro benchmark, GPT-5.4 demonstrated 81.2% success rate without the use of external tools. This is higher than Gemini 3.1 Pro, which demonstrated 80.5%.
Read also: Trump vs. Claude AI: How the AI Debate Is Unfolding in the U.S.
One of the constant complaints about GPT models was the insufficient quality of the documents and presentations they generated. OpenAI has worked hard to address this shortcoming: according to the company, GPT-5.4 produces more aesthetically pleasing materials, offers greater visual variety, and makes better use of image generation capabilities. At the same time, the company emphasizes that the new model is the most efficient of all its reasoning models in terms of token utilization.
According to the benchmark comparison tables, the GPT-5.4 series outperforms competing models from Anthropic and Google in most key indicators.
For developers, the Codex environment has a new /fast mode that provides up to 1.5 times faster token processing when using GPT-5.4 without losing the quality of the results. In the API, instead of preloading all tool definitions in the request, the new model receives a lightweight list of available tools along with a search function.
GPT-5.4 Thinking is already available for ChatGPT Plus, Team, and Pro users, replacing GPT-5.2 Thinking. At the same time, GPT-5.2 Thinking will remain on the list of outdated models for another three months. GPT-5.4 Pro is intended exclusively for subscribers of ChatGPT Pro and Enterprise plans. As for API prices: GPT-5.4 costs $2.50 per million incoming tokens and $15 per million outgoing tokens. The GPT-5.4-pro version is much more expensive – $30 for 1 million incoming tokens and $180 for 1 million outgoing tokens.
Read also:
- Camera, biometrics, and ChatGPT: What we know about the first OpenAI device
- Report from the Xiaomi Booth at #MWC2026: Leica-Equipped Flagships, AI Tablets, and 1900 HP in the Vision GT