OpenAI Launches GPT-5.4 With Pro and Thinking Variants
Key Facts
- What: OpenAI released GPT-5.4, a new frontier model optimized for professional work, available in standard, GPT-5.4 Thinking (reasoning), and GPT-5.4 Pro (high-performance) versions.
- When: Released Thursday, March 5, 2026.
- Context Window: API version supports up to 1 million tokens, the largest from OpenAI to date.
- Benchmarks: Record scores on OSWorld-Verified, WebArena Verified, 83% on OpenAI’s GDPval knowledge work test, and leadership on Mercor’s APEX-Agents benchmark for law and finance.
- Availability: GPT-5.4 Thinking available to Plus, Team, and Pro users; GPT-5.4 Pro limited to Pro and Enterprise plans, with API access for both.
Lead paragraph
OpenAI on Thursday released GPT-5.4, its latest foundation model described as the company’s “most capable and efficient frontier model for professional work.” The model is offered in three variants — a standard version, a reasoning-focused GPT-5.4 Thinking edition, and a high-performance GPT-5.4 Pro version — and brings significant gains in benchmark performance, token efficiency, and tool-calling capabilities. The launch underscores OpenAI’s continued push to deliver more reliable AI systems for complex professional tasks in law, finance, and knowledge work.
Model Variants and Capabilities
According to TechCrunch, GPT-5.4 builds on previous releases with improved reasoning and efficiency. The Thinking variant is designed for multi-step reasoning tasks, while the Pro version targets maximum performance on demanding workloads. OpenAI highlighted the model’s ability to create long-horizon deliverables such as slide decks, financial models, and legal analysis.
The API version of GPT-5.4 supports context windows of up to 1 million tokens. OpenAI also reported substantial gains in token efficiency, stating the new model can solve the same problems using significantly fewer tokens than its predecessor, GPT-5.2.
Benchmark Performance
GPT-5.4 posted record results across several evaluations. It achieved new highs on computer use benchmarks OSWorld-Verified and WebArena Verified. The model scored 83% on OpenAI’s GDPval test, which measures performance on knowledge work tasks.
On Mercor’s APEX-Agents benchmark — focused on professional skills in law and finance — GPT-5.4 took the lead, according to Mercor CEO Brendan Foody. “GPT-5.4 excels at creating long-horizon deliverables such as slide decks, financial models, and legal analysis,” Foody said in a statement. “It delivers top performance while running faster and at a lower cost than competitive frontier models.”
Additional reporting notes that GPT-5.4 Pro achieved a state-of-the-art 89.3% on the BrowseComp benchmark, which tests an AI agent’s ability to persistently browse the web for hard-to-locate information — a 17 percentage point improvement over GPT-5.2 in related testing.
Safety and Technical Improvements
OpenAI emphasized continued progress on reducing hallucinations and factual errors. The company reported that GPT-5.4 is 33% less likely to make errors in individual claims compared to GPT-5.2, with overall responses 18% less likely to contain errors.
The launch introduces a new Tool Search system for API users. Previously, tool definitions were included in system prompts, which consumed significant tokens when many tools were available. The new approach lets models look up tool definitions as needed, resulting in faster and cheaper requests for complex systems.
OpenAI also added a new safety evaluation focused on chain-of-thought (CoT) monitoring. AI safety researchers have expressed concern that reasoning models could misrepresent their internal thought processes. Testing of GPT-5.4 Thinking suggests the model is less likely to engage in deception, indicating it “lacks the ability to hide its reasoning and that CoT monitoring remains an effective safety tool,” according to the company.
Availability and Access
The model is rolling out across ChatGPT, the API, and Codex. GPT-5.4 Thinking is available to ChatGPT Plus, Team, and Pro users, while Enterprise and Edu customers require admin enablement. GPT-5.4 Pro is restricted to Pro and Enterprise plans and available through the API.
Impact
The release positions OpenAI to better serve professional users who require reliable performance on complex, multi-step tasks. Improved token efficiency and the new Tool Search system should reduce operational costs for developers building agentic workflows. Stronger benchmark results in computer use, web browsing, and professional domains may accelerate adoption in legal, financial, and enterprise settings.
What's Next
OpenAI has not announced a specific timeline for further GPT-5.x releases or additional variants. The company is expected to continue refining safety evaluations and expanding context windows as competition in frontier models intensifies.
Sources
- OpenAI launches GPT-5.4 with Pro and Thinking versions | TechCrunch
- Introducing GPT-5.4 | OpenAI
- OpenAI launches GPT-5.4 Thinking and Pro - The New Stack
- OpenAI GPT-5.4 Thinking Released | Business Standard
- OpenAI, in Desperate Need of a Win, Launches GPT-5.4 | Gizmodo
All technical specifications, pricing, and benchmark data in this article are sourced directly from official announcements. Competitor comparisons use publicly available data at time of publication. We update our coverage as new information becomes available.