OpenAI has released GPT-5.4 across ChatGPT, Codex and the API, claiming it as a single model that combines recent gains in reasoning, coding and agentic workflows.
The firm also launched ChatGPT for Excel in beta and rolled out new financial data integrations in ChatGPT, tying the model update to a more explicit enterprise push into spreadsheet-heavy and research-heavy finance work.
In Codex and the API, GPT-5.4 is OpenAI’s first general-purpose model with native computer-use capabilities, allowing agents to operate computers and carry out multi-application workflows.
The same model supports up to 1 million tokens of context, up from the 400,000-token limit on GPT-5 and GPT-5.3-Codex in the API.
Accuracy and efficiency claims for production use
In its launch post, the company said GPT-5.4 is 33% less likely than GPT-5.2 to make false individual claims and 18% less likely to produce a response containing any error.
In the API, OpenAI said a new tool search feature reduced total token usage by 47% at the same accuracy level in its test on 250 MCP Atlas tasks with 36 MCP servers enabled, a result the company presented as a way to lower cost and latency in tool-heavy agent systems.
The finance push
OpenAI said GPT-5.4 was improved on real-world tasks including financial modeling, scenario analysis, data extraction and long-form research.
On OpenAI’s internal benchmark for spreadsheet modeling tasks that a junior investment banking analyst might do, GPT-5.4 scored 87.3%, compared with 68.4% for GPT-5.2, according to the company. OpenAI paired that claim with ChatGPT for Excel, which it said can build, update and analyze spreadsheet models directly inside workbooks.
That spreadsheet product is rolling out in beta to ChatGPT Business, Enterprise, Edu, Teachers, Pro and Plus users in the U.S., Canada and Australia, while ChatGPT for Google Sheets is still listed as coming soon.
OpenAI also said enterprise, edu and teacher workspaces will have the feature off by default, with admins able to enable access for specific users through custom roles and group permissions. Alongside Excel, OpenAI said March 5 integrations include Moody’s, Dow Jones Factiva, MSCI, Third Bridge and MT Newswire, with FactSet coming soon.
Security posture and competitive context
OpenAI’s finance launch said ChatGPT Enterprise includes RBAC, SAML SSO, SCIM and audit logs, along with TLS 1.2+ in transit, AES-256 at rest, data residency controls and a default policy under which enterprise data is not used to train its models.
OpenAI’s safety documentation also said GPT-5.4 Thinking is the first general-purpose model to include mitigations for “High capability in Cybersecurity.”
That positioning puts OpenAI more directly into a market Anthropic and Google have already been building toward. Anthropic’s Claude for Financial Services, introduced in July 2025, combines Claude, Claude Code, pre-built MCP connectors and direct links to source materials for verification.
Google has positioned Gemini Enterprise as a platform that integrates with organizational data and supports unified governance for agents, while Vertex AI remains its fully managed development platform for building and deploying AI applications.
Migration path and API pricing
For existing OpenAI customers, the migration path is now defined. GPT-5.4 Thinking is available in ChatGPT for Plus, Team and Pro users, replacing GPT-5.2 Thinking, while GPT-5.2 Thinking remains in the legacy model picker for paid users until June 5, 2026.
GPT-5.4 Pro is available to Pro and Enterprise plans. In the API, OpenAI says prompts above 272,000 input tokens are billed at double input rates and 1.5x output rates for the full session.