OpenAI Unveils Enterprise-Grade GPT-5.4
OpenAI just released GPT-5.4, its most advanced model designed for enterprise work. The new model reportedly outperforms humans by 83% on professional coding and writing benchmarks, features a 1M token context window, and can natively use computer files and tools. The launch is seen as a direct challenge to rival Anthropic for high-value corporate contracts.
The launch of GPT-5.4 is part of a broader strategy at OpenAI to push deeper into high-value enterprise workflows, particularly in sectors like finance. The model's release was bundled with a beta of ChatGPT for Excel and new integrations with financial data providers including Moody's, Dow Jones Factiva, and MSCI. This move is designed to embed OpenAI's technology directly into the spreadsheet and research-heavy processes that define many corporate environments. A key technical advance in GPT-5.4 is its "agentic" ability to natively control a computer, including using a mouse and keyboard to interact with websites and applications. This allows it to perform complex, multi-step tasks like reading an email, extracting an attachment, and logging the data in a spreadsheet without human intervention. On the OSWorld-Verified benchmark, which measures the ability to navigate a desktop environment, GPT-5.4 scores 75.0%, surpassing the human baseline of 72.4%. The model's 1 million token context window is a significant expansion from previous versions like GPT-4, which topped out at 128,000 tokens. This larger "working memory" allows the AI to process and analyze the equivalent of thousands of pages of documents, such as entire codebases or years of financial records, in a single pass. However, to manage costs, OpenAI bills prompts that exceed 272,000 input tokens at double the standard rate. This release intensifies the rivalry with Anthropic, which has been successfully targeting the enterprise market by emphasizing AI safety and reliability. Anthropic recently launched its own "Claude Enterprise" plan and a "Cowork" platform with partners like PwC to build specialized AI agents for departments like finance and HR. Anthropic has captured an estimated 32% of the enterprise AI market by focusing on corporate contracts over a consumer-facing, ad-supported model. The competition between the two firms extends to government contracts. Just prior to the GPT-5.4 launch, the Pentagon blacklisted Anthropic, designating it a "supply chain risk" after the company refused to remove safeguards that would prevent its AI from being used for mass surveillance or autonomous weapons. OpenAI then quickly signed a deal to provide its models for classified use by the Department of Defense, agreeing to a compromise that references existing laws as guardrails.
