OpenAI Launches GPT-5.4 with Pro and Thinking Versions
SAN FRANCISCO — OpenAI on Thursday released GPT-5.4, a new flagship model described as “our most capable and efficient frontier model for professional work.” The release includes two specialized variants — GPT-5.4 Thinking, a reasoning-focused model, and GPT-5.4 Pro — now available across ChatGPT, the company’s API and Codex.
The launch marks OpenAI’s latest effort to advance AI capabilities for complex professional tasks, with notable gains in reasoning, coding and agent-driven workflows. GPT-5.4 Thinking and GPT-5.4 Pro were rolled out starting Thursday to select ChatGPT subscribers and through the company’s developer platforms.
According to OpenAI’s announcement, the standard GPT-5.4 model emphasizes both capability and efficiency for professional use cases. The Thinking version is designed to deliver stronger step-by-step reasoning, while the Pro variant targets high-performance scenarios requiring maximum intelligence.
Availability and Access Tiers
GPT-5.4 Thinking is available to ChatGPT Plus, Team and Pro users. Enterprise and Education customers can access it once enabled by their administrators. GPT-5.4 Pro is more restricted, available only to users on Pro and Enterprise plans, as well as through the API for qualified developers.
The models have been integrated into ChatGPT, the OpenAI API and Codex, OpenAI’s AI-powered coding assistant. This broad rollout allows both individual professionals and organizations to begin testing the new capabilities immediately.
Performance Gains on Agent Benchmarks
OpenAI highlighted significant improvements on agent-oriented benchmarks. On BrowseComp — a test measuring how effectively AI agents can persistently browse the web to locate difficult-to-find information — GPT-5.4 shows a 17-point gain over GPT-5.2. GPT-5.4 Pro establishes a new state-of-the-art score of 89.3% on the same benchmark.
These results suggest the new models are particularly well-suited for research, data gathering and multi-step workflows that require reliable web navigation and information synthesis. The company positioned GPT-5.4 as a practical upgrade for users tackling sophisticated professional tasks rather than a pure research breakthrough.
Competitive Context
The release arrives as OpenAI faces growing competition from rivals including Anthropic, Google DeepMind and xAI. Industry observers have noted OpenAI’s need to demonstrate continued leadership in frontier model development following a series of high-profile product updates and internal challenges in 2025 and early 2026.
By offering both a “Thinking” reasoning model and a high-performance “Pro” variant, OpenAI is following a pattern seen in previous releases where specialized versions cater to different user needs and compute budgets.
Impact on Developers and Users
For developers, the API availability of both GPT-5.4 Thinking and GPT-5.4 Pro opens new possibilities for building more capable AI agents and professional tools. The improved BrowseComp performance is expected to benefit applications that require autonomous research, competitive intelligence and complex information retrieval.
Enterprise users gain access to models specifically optimized for professional workflows, potentially improving productivity in coding, analysis and knowledge work. However, the tiered availability means many individual users on lower-tier plans will not immediately access the most powerful Pro version.
What’s Next
OpenAI has not yet disclosed detailed model size, training data or comprehensive benchmark suites beyond the BrowseComp results. Further technical documentation and additional evaluation results are expected in the coming weeks.
The company indicated that GPT-5.4 represents an incremental but meaningful step in its frontier model roadmap. No specific timeline was provided for a potential GPT-5.5 or next major foundation model release.
Users with access to eligible plans can begin using GPT-5.4 variants in ChatGPT immediately, while developers should consult the updated API documentation for integration details.
This article is based on OpenAI’s official announcement and contemporaneous reporting from TechCrunch, The New Stack, Business Standard and Gizmodo.