OpenAI Launches GPT-5.4: Million-Token Context, Computer Use, and a New Benchmark Standard
OpenAI's latest frontier model GPT-5.4 introduces a one-million-token context window, native computer-use capabilities, and outperforms industry professionals in 83% of benchmark comparisons — marking the most aggressive model iteration cycle in the company's history.
AI Summary
OpenAI GPT-5.4 launch million-token context computer-use capabilities benchmarks GPT-5.3 Codex API ChatGPT Plus Pro reasoning coding agentic AI GPT-5.2 retirement frontier model iteration cycle March 2026
OpenAI has released GPT-5.4, its most capable frontier model to date, introducing a set of features that significantly expand the boundaries of what large language models can accomplish. Available to ChatGPT Plus, Team, and Pro users, as well as developers through the API and Codex platform, the model represents the culmination of a rapid iteration cycle that has seen five major GPT-5 variants released since the series launched in August 2025.
A Million Tokens of Context
The headline feature of GPT-5.4 is native support for up to one million tokens of context in the API and Codex — a substantial leap that enables the model to process entire codebases, lengthy legal documents, or comprehensive datasets in a single conversation. This capability positions GPT-5.4 as a direct competitor to Anthropic's Claude Opus 4.6, which introduced its own million-token context window in February 2026.
Computer-Use Capabilities
GPT-5.4 introduces native computer-use capabilities, allowing the model to operate across tools and software environments by recognizing interface elements from screenshots and performing mouse and keyboard operations. The GPT-5.4 Thinking variant in ChatGPT can outline its reasoning process, giving users transparency into the model's decision-making and the ability to adjust instructions in real time.
Benchmark Performance
In testing, GPT-5.4 matched or exceeded industry professionals in 83% of GDPval benchmark comparisons and demonstrated significantly higher scores in coding ability assessments compared to previous models. OpenAI reports improved factual accuracy relative to GPT-5.2, with reduced hallucination rates across a range of tasks including spreadsheets, presentations, and document analysis.
The Iteration Arms Race
The release of GPT-5.4 underscores an industry-wide acceleration in model development cycles. Major AI labs are now shipping updates every two to three weeks, driving capabilities higher while reducing inference costs. The frontier AI cost curve has dropped tenfold compared to 2025, democratizing access to capabilities that were exclusive to well-funded enterprises just a year ago.
OpenAI plans to gradually retire GPT-5.2 Thinking in the coming months as users transition to the newer model. GPT-5.4 Pro, offering maximum performance for complex professional tasks, is available to ChatGPT Pro subscribers.
Competitive Landscape
The release comes amid intense competition. Google's Gemini 3.1 Pro dominates 13 out of 16 major benchmarks. Anthropic's Claude Opus 4.6, released in February, features enhanced coding and financial analysis capabilities. xAI's Grok 4.20 introduced a four-agent architecture, and DeepSeek V4 launched with one trillion parameters and 40% memory reduction through architectural innovations. China's MiniMax M2.5 is positioned as an affordable rival to Claude Opus 4.6, and Alibaba has entered the space with its Qwen 3.5 series.
For developers, GPT-5.4 is available immediately through the OpenAI API. Enterprise customers can access the model through Azure OpenAI Service with the same million-token context window and computer-use capabilities.