OpenAI Unveils GPT-5.4: A New Frontier in AI Models

On March 5, 2026, OpenAI introduced GPT-5.4, a new foundation model designed for professional applications. This release includes two specialized versions: GPT-5.4 Thinking, optimized for complex reasoning tasks, and GPT-5.4 Pro, tailored for high-performance scenarios. The API version of GPT-5.4 boasts a context window of up to 1 million tokens, the largest offered by OpenAI to date. The company claims that GPT-5.4 demonstrates improved token efficiency, solving problems with fewer tokens compared to its predecessor, GPT-5.2. Notably, it achieved record scores in various benchmarks, including an 83% on OpenAI’s GDPval test for knowledge work tasks and top performance in Mercor’s APEX-Agents benchmark, which evaluates skills in law and finance.

TechCrunchCNET

OpenAI has emphasized that GPT-5.4 is designed to minimize hallucinations and factual inaccuracies. The model is reported to be 33% less likely to make errors in individual claims and 18% less likely to contain errors overall compared to GPT-5.2. Additionally, the introduction of a new system called Tool Search allows the model to access tool definitions more efficiently, resulting in faster and more cost-effective requests. OpenAI has also implemented a safety evaluation to assess the model's chain-of-thought, indicating that the Thinking version is less likely to misrepresent its reasoning.

TechCrunch

The launch of GPT-5.4 comes amid increasing competition from Anthropic's Claude, which has gained popularity in app stores. OpenAI's new models are positioned to attract enterprise users, particularly those engaged in coding and managing AI agents. The rivalry between OpenAI and Anthropic has intensified, especially following Anthropic's refusal to allow the U.S. government to use its AI for surveillance and military applications. OpenAI has stepped into this space, having previously secured a $200 million contract with the U.S. Department of War in 2025, with CEO Sam Altman assuring that safeguards will prevent the use of its technology by intelligence agencies.

CNET

Despite the advancements, there are ongoing concerns regarding the ethical implications of AI technology in government use. Critics highlight the need for transparency in how AI models are deployed, particularly in military contexts. The debate surrounding AI's role in surveillance and autonomous weapon systems remains contentious, with calls for clearer regulations and oversight.

CNET

• “[GPT-5.4] excels at creating long-horizon deliverables such as slide decks, financial models, and legal analysis,” — Brendan Foody, CEO of Mercor

TechCrunch

• “most factual model yet,” — OpenAI Statement

CNET

• “OpenAI stepped into that void, with CEO Sam Altman clarifying this week that it would implement safeguards and wouldn't be made available to intelligence agencies like the NSA.” — Sam Altman, CEO of OpenAI

CNET

The introduction of GPT-5.4 marks a significant step in OpenAI's evolution, reflecting both technological advancements and the complexities of navigating ethical considerations in AI deployment.

CNET