OpenAI Launches GPT-5.4 With Native Computer Control and 1M Token Context [Video]

OpenAI has released GPT-5.4, a major update that brings native computer-use capabilities and a massive context window to its artificial intelligence lineup. Available today across ChatGPT, the API, and Codex, the new model is built to handle complex professional workloads with far less manual setup. The company also introduced a higher-tier GPT-5.4 Pro designed for the most demanding enterprise tasks.



The standout feature of this release is direct computer control. GPT-5.4 is the first general-purpose model from OpenAI capable of navigating a desktop environment. It can issue keyboard commands and control a mouse in response to reading screenshots. For Mac developers, this arrives at a highly relevant time. OpenAI recently launched a dedicated Codex app for macOS, and Apple just integrated official support for agentic AI coding in Xcode 26.3. Developers can now steer GPT-5.4 to handle multi-step coding tasks across different local applications, with configurable safety guardrails.


To support these heavier workloads, the model now supports context windows of up to one million tokens. This allows it to ingest entire codebases, lengthy contracts, or massive API documentation libraries in a single prompt. For those writing code, a new fast mode in Codex is designed to keep developers moving quickly during debugging. There is also an experimental interactive skill that allows Codex to visually debug web and Electron apps as they are being built.

Visual perception received a significant upgrade as well. The model introduces a new image input tier that supports full-fidelity perception up to 10.24 million pixels. This allows the AI to better understand dense, high-resolution images, improving document parsing and click accuracy when operating software.

Inside the standard ChatGPT interface, the model operates as GPT-5.4 Thinking. It now generates an upfront plan before writing out a full response to a complex prompt. If you see the AI going down the wrong path, you can interrupt and adjust its direction mid-response, saving you from having to start over. OpenAI says the model is also more persistent during deep web research, maintaining its context better when digging through multiple sources to pull out highly specific information.

Accuracy also improves. Building on the refinement seen in the recent GPT-5.3 Instant release, full responses from GPT-5.4 are 18 percent less likely to contain factual errors compared to the previous generation. This makes it significantly more reliable for data-heavy tasks. OpenAI also launched a dedicated ChatGPT for Excel add-in for enterprise customers today.


For API developers, a new tool search function drastically cuts costs. Instead of loading thousands of tokens worth of external tool definitions into every prompt, GPT-5.4 receives a lightweight index. It only pulls the full definition into the context window when it actually needs to use the specific tool. OpenAI says this approach can cut token usage by nearly half in complex, multi-step workflows.

GPT-5.4 Thinking is rolling out today for ChatGPT Plus, Team, and Pro subscribers, replacing the older 5.2 version. The previous model will remain accessible in a legacy menu for three months before being retired in June. API pricing for the standard GPT-5.4 model sits at $2.50 per million input tokens and $15 per million output tokens. The more powerful Pro variant commands a premium, priced at $30 for input and $180 for output tokens.