OpenAI Unveils GPT-5.4 Featuring Integrated Financial Tools for Spreadsheets and Investment Analysis

Posted on by

TLDR

  • On March 5, 2026, OpenAI introduced GPT-5.4, its most advanced model for professional tasks.
  • The new model features financial tools integrated with the FactSet and Third Bridge data platforms.
  • It can be accessed directly within Microsoft Excel and Google Sheets.
  • GPT-5.4 directly rivals Anthropic’s Claude for Financial Services.
  • Access begins today for ChatGPT Plus, Team, and Pro subscribers.

OpenAI introduced its newest AI model, GPT-5.4, on Thursday, March 5, 2026. This launch incorporates a suite of new tools targeted at finance experts.

The model produces spreadsheets, documents, and presentations with less need for iterative revisions. It also performs web searches to collect data and formulate responses to intricate queries.

OpenAI incorporated financial tools in this release that link to data providers FactSet Research Systems and Third Bridge. These connections are intended for users conducting financial analysis and creating investment memorandums.

The model is now accessible directly within Excel and Google Sheets. A ChatGPT add-in for Excel was also released concurrently.

This launch increases competitive pressure on Anthropic, which debuted Claude for Financial Services the previous year. The two firms are vying to secure business clients prepared to invest in AI solutions.

GPT-5.4 Performance and Benchmarks

In an internal evaluation of investment banking spreadsheet work, GPT-5.4 achieved a score of 87.3%, a rise from the 68.4% scored by the prior GPT-5.2 model. Human evaluators favored its presentations over those from GPT-5.2 68% of the time.

On the GDPval benchmark, which assesses AI performance across 44 professions, GPT-5.4 equaled or surpassed industry experts in 83% of comparisons. GPT-5.2 attained 70.9% on this same test.

GPT-5.4 also recorded a score of 75% on OSWorld-Verified, a test that evaluates a model’s ability to operate a desktop using screenshots and mouse commands. Human performance on this test is 72.4%.

OpenAI describes this model as its most accurate to date. Individual statements are 33% less probable to be incorrect relative to GPT-5.2.

Availability and Pricing

GPT-5.4 is being released to ChatGPT Plus, Team, and Pro subscribers today as GPT-5.4 Thinking. The older GPT-5.2 Thinking model will remain accessible for three months before being discontinued on June 5, 2026.

Users on Enterprise and Edu plans can activate early access via administrator settings. GPT-5.4 Pro is offered to Pro and Enterprise plan customers.

API pricing begins at $2.50 per million input tokens and $15 per million output tokens. This represents an increase from GPT-5.2, which was priced at $1.75 and $14 per million tokens, respectively.

In Codex, the model accommodates context windows of up to one million tokens. Standard API calls support 272,000 tokens, and requests exceeding this limit are charged at double the standard rate.

According to OpenAI, GPT-5.4 uses tokens more efficiently than GPT-5.2, a factor the company states helps lower overall expenses despite the higher per-token cost.

The model is also accessible via the company’s AI coding assistant, Codex, and through the standard API under the identifier gpt-5.4.