Yesterday, OpenAI released the latest update to its AI, GPT-5.4, across ChatGPT as GPT-5.4 Thinking, the API, and Codex. On the same day, the company launched ChatGPT for Excel in beta. Both are available now.
GPT-5.4 is the first OpenAI general-purpose model with native computer-use capability built directly into it. The model can read screenshots, move a cursor, and use a keyboard to operate a desktop environment without a separate tool layer. On OSWorld-Verified, the benchmark that tests AI capabilities, GPT-5.4 scored 75.0%. The human baseline on that same test is 72.4%. GPT-5.2 scored 47.3%.
Alongside the model, OpenAI launched ChatGPT for Excel in beta — an add-on that embeds ChatGPT directly inside a workbook. Users describe their needs in plain language, and the model builds or updates live Excel models using the workbook's existing formulas and structure. It links answers to the exact cells it references and updates, explains what it is doing as it works, and asks for permission before making any changes.
Damilare Odedina
On OpenAI's internal investment banking benchmark — building a three-statement financial model with proper formatting and citations — GPT-5.4 Thinking scored 87.3%, up from 68.4% with GPT-5.2 and 43.7% with the original GPT-5.
GPT-5.4 also incorporates the coding capabilities of GPT-5.3-Codex. On SWE-Bench Pro it scores 57.7%, ahead of GPT-5.3-Codex at 56.8% and GPT-5.2 at 55.6%. A “/fast” mode inside Codex delivers up to 1.5x faster token output from the same model. Per OpenAI's published benchmarks, GPT-5.4 uses fewer reasoning tokens than GPT-5.2 to reach the same output quality, translating to reduced token usage and faster speeds.
On factual accuracy, OpenAI tested GPT-5.4 against a set of real user-flagged prompts where errors had previously been reported. Individual claims came in 33% less likely to be false, and full responses 18% less likely to contain errors compared to GPT-5.2.
Matt Shumer, founder and CEO of OthersideAI, tested GPT-5.4 for a week before the public launch and posted on X: "Even in standard mode, GPT-5.4 is better than previous models in Pro mode... Coding capabilities are ridiculous. It's essentially flawless. Inside Codex, it's insanely reliable."
He flagged three gaps—frontend design quality, occasional missed real-world context, and tasks stopping short inside his tool, OpenClaw. OpenAI CEO Sam Altman replied to the post: "We will be able to fix these three things."
Who gets it, what it costs, and what's coming next
ChatGPT for Excel in beta is available now to Plus, Pro, Business, Enterprise, Edu, and Teachers users in the U.S., Canada, and Australia. Enterprise, Edu, and Teacher workspaces have it off by default — admins need to enable it. New financial data integrations are also live inside ChatGPT from Moody's, Dow Jones Factiva, MSCI, Third Bridge, and MT Newswire, with FactSet listed as coming soon. ChatGPT for Google Sheets is listed as coming soon.
In ChatGPT, GPT-5.4 Thinking is available now for Plus, Team, and Pro users, replacing GPT-5.2 Thinking. GPT-5.2 Thinking stays in the Legacy section of the model picker until it retires on June 5, 2026. GPT-5.4 Pro is available on Pro and Enterprise plans. Enterprise and Edu admins can enable early access via admin settings.
In the API, gpt-5.4 and gpt-5.4-pro are available, with a context window of up to 1 million tokens. GPT-5.4 is priced at $2.50 per million input tokens and $15 per million output tokens, compared to GPT-5.2's $1.75 input and $14 output. GPT-5.4 Pro runs at $30 per million input and $180 per million output. Batch and Flex processing are available at half the standard rate. Priority processing is available at twice the standard rate.
Damilare Odedina
