OpenAI has launched GPT-5.4, including variants Thinking and Pro, aimed at improving agentic tasks and knowledge work. The update features enhanced computer-use capabilities and reduced factual errors, amid competition from Anthropic following a US defense deal controversy. The models are available immediately to paid users and developers.
On March 5, 2026, OpenAI released GPT-5.4, a new model family including GPT-5.4 Thinking and GPT-5.4 Pro, as part of its accelerated release schedule. This follows the launch of GPT-5.3 Instant earlier in the week. The company describes GPT-5.4 as its first model explicitly designed for computer-use tasks, enabling it to issue keyboard and mouse inputs based on desktop or application screenshots. It supports multi-app operations and is optimized for agentic activities, such as coding, data analysis, and overseeing AI agents, with improved efficiency in computing power and costs.
GPT-5.4 Thinking, now the default for ChatGPT's Thinking mode among paid subscribers, presents reasoning upfront and allows users to adjust prompts mid-process. It maintains better context for long reasoning tasks and enhances web research for specific queries. The API version offers a 1 million token context window and is more token-efficient, though input tokens cost $2.50 per million, up from $1.75 for GPT-5.2. Visual understanding has advanced, handling images up to 10.24 million pixels or 6,000-pixel dimensions. OpenAI claims responses are 18 percent less likely to contain factual errors compared to GPT-5.2, with individual claims 33 percent less likely to be false.
The release occurs amid user shifts to competitors like Anthropic's Claude, driven by OpenAI's $200 million deal with the US Department of Defense announced in 2025. This followed Anthropic's refusal to allow its models for surveillance or autonomous weapons, leading to a feud. OpenAI states it implements safeguards and excludes use by agencies like the NSA. Anthropic responded by extending its memory feature to free users and reported its largest sign-up day on March 2, 2026. The models are accessible via ChatGPT web and apps, Codex, and API for Plus, Team, Pro, Edu, and Enterprise users.
OpenAI's focus on professional tools aligns with its strategy, generating about $25 billion in annualized revenue but remaining unprofitable amid $1.4 trillion in data center commitments.
