TLDR
- GPT-5.4 supports up to 1M tokens of context
-
Model integrates GPT-5.3 Codex coding features
-
GDPval benchmark score reaches 83%
-
GPT-5.4 Pro priced at $30 and $180 per million tokens
OpenAI has introduced GPT-5.4 as its newest frontier AI model. The system is designed for professional workloads and advanced enterprise use.
The company is deploying the model across ChatGPT, the API, and Codex. A higher performance GPT-5.4 Pro version is also available for complex tasks.
OpenAI stated that GPT-5.4 combines advanced reasoning and coding in one model. The release reflects the company’s focus on enterprise and developer tools.
Enhanced Reasoning and Coding Integration
GPT-5.4 integrates the coding capabilities first introduced in GPT-5.3 Codex. It improves performance across software development and document workflows. The model supports spreadsheets, presentations, and structured data tasks. It can also outline its reasoning plan in advance within ChatGPT.
GPT-5.4 Thinking and GPT-5.4 Pro are rolling out now in ChatGPT.
GPT-5.4 is also now available in the API and Codex.
GPT-5.4 brings our advances in reasoning, coding, and agentic workflows into one frontier model. pic.twitter.com/1hy6xXLAmJ
— OpenAI (@OpenAI) March 5, 2026
This feature allows users to guide the response while it is still in progress. OpenAI said this reduces wasted tokens during long reasoning chains. The model supports a context window of up to one million tokens. OpenAI has removed the beta label for this feature in the API.
Codex also supports the extended token window. However, requests above 272,000 tokens count at double the usage rate.
Native Computer Use and Tool Search
GPT-5.4 introduces native computer use features. AI agents can interact with operating systems, websites, and applications using keyboard and mouse inputs.
Developers can automate multi-step workflows across different software tools. The system can also select tools dynamically through a feature called tool search.
Tool search enables the model to locate and use external tools without loading all definitions into memory. This approach reduces token usage in complex environments.
OpenAI reported that GPT-5.4 responses are 18% less likely to contain errors than GPT-5.2. Individual claims are 33% less likely to be false. The company said GPT-5.4 is its most token-efficient reasoning model to date. It uses fewer tokens to complete many reasoning tasks.
Benchmarks and Enterprise Performance
On the GDPval benchmark, GPT-5.4 scored 83%. This benchmark measures performance across 44 occupations in professional knowledge work.
OpenAI said the model matched or exceeded industry professionals in 83% of comparisons. Anthropic’s Opus 4.6 scored 79.5% on the same test. On the FrontierMath benchmark, GPT-5.4 Pro scored 38% on advanced math problems. The Thinking version scored 27.1%.
The model also performed strongly on SWE-Bench Pro coding tests. It outperformed earlier OpenAI models and several competitors.
In internal testing, human raters preferred GPT-5.4 presentations 68% of the time over GPT-5.2. The model also scored 87.5% on a spreadsheet task modeled on junior banking workflows.
Pricing and Availability
GPT-5.4 is available in Thinking and Pro versions. GPT-5.4 Thinking costs $2.50 per million input tokens and $15 per million output tokens.
GPT-5.4 Pro costs $30 per million input tokens and $180 per million output tokens. It is OpenAI’s highest priced model to date. The Thinking version is available to Plus, Team, and Pro users in ChatGPT. The Pro version is available to Pro and Enterprise users.
Developers can access the models through the API under gpt-5.4 and gpt-5.3-pro. The rollout expands OpenAI’s AI infrastructure across consumer and enterprise platforms.
