GPT-5.4 is now available

We’re excited to announce the release of GPT-5.4—the most capable and efficient model for professional work across the API, Codex, and ChatGPT. It completes complex tasks with less back-and-forth while using significantly fewer tokens than GPT-5.2.

GPT-5.4 combines the industry-leading coding capabilities of GPT-5.3-Codex with stronger performance on long-running tasks, improved front-end UI generation, and built-in computer use (CUA). It also works seamlessly across tools—analyzing and creating spreadsheets, presentations, and documents—making it an ideal model for domains like financial services, data analysis, and customer support.

To get the most out of GPT-5.4, check out our API prompting guide and migration guide.

Build with GPT 5.4

Here’s how GPT-5.4 performs across key capabilities:

Knowledge work: GPT-5.4 delivers even more consistent and polished results on real-world tasks that matter to professionals. On GDPval, GPT-5.4 achieves a new state of the art, matching or exceeding industry professionals in 83% of comparisons, compared to 71% for GPT-5.2.

Computer use and vision: GPT-5.4 introduces native computer use and achieves a state-of-the-art 75% on OSWorld-Verified (vs. 37.9% on GPT-5.2), making it our best model for building computer-operating agents. GPT-5.4’s vision capabilities are also greatly improved, particularly with the new original detail parameter in the API.

Coding: GPT-5.4 is our best coding model yet—matching or outperforming GPT‑5.3‑Codex on SWE-Bench Pro with lower latency. It also features a larger, 1M-tokens context window, enabling longer-running tasks where the model can use tools, iterate, and verify its work.

Tool use: GPT-5.4, significantly improves how models work with external tools. It introduces tool search, allowing models to dynamically load tools into context without breaking the cache. It also improves tool calling performance across reasoning efforts, particularly for low-latency cases where reasoning is disabled.

Improved web search: GPT-5.4 is better at agentic web search. On BrowseComp, a measurement of how well AI agents can persistently browse the web to find hard-to-locate information, GPT-5.4 leaps 17% points over GPT-5.2, and GPT-5.4 Pro sets a new state of the art of 89.3%.

Build with GPT 5.4

—The OpenAI Team

Introducing GPT-5.4

AI Analysis

Extracted Information