TLDR
- OpenAI unveiled GPT-5.4 on March 5, 2026, positioning it as its most advanced professional-grade model
- Financial capabilities include direct integration with FactSet Research Systems and Third Bridge databases
- Direct accessibility within Microsoft Excel and Google Sheets environments
- Direct competition with Anthropic’s Claude for Financial Services offering
- Accessible to ChatGPT Plus, Team, and Pro tier subscribers immediately
On Thursday, March 5, 2026, OpenAI introduced GPT-5.4, its newest artificial intelligence model specifically enhanced for professional applications. The update brings specialized capabilities targeting the finance sector.
This iteration streamlines the creation of spreadsheets, documentation, and slide decks, requiring fewer conversational iterations. The system performs web searches to compile data and deliver responses to sophisticated queries.
The financial functionality connects directly with FactSet Research Systems and Third Bridge, two prominent data service providers. These connections support professionals conducting market research and preparing investment documentation.
Direct integration within Microsoft Excel and Google Sheets environments is now possible. OpenAI simultaneously released a ChatGPT add-in specifically designed for Excel.
This launch intensifies competition with Anthropic, whose Claude for Financial Services debuted previously. Both organizations are vying for corporate clients prepared to invest in artificial intelligence solutions.
Performance Metrics and Testing Results
During internal evaluation focused on investment banking spreadsheet operations, GPT-5.4 achieved an 87.3% success rate, significantly improved from GPT-5.2’s 68.4% performance. In presentation quality assessments, human evaluators favored GPT-5.4 output 68% of the time versus GPT-5.2.
Using GDPval, a benchmark spanning 44 professional occupations, GPT-5.4 equaled or exceeded human expert performance in 83% of test scenarios. The prior GPT-5.2 version achieved 70.9% on identical assessments.
On OSWorld-Verified, which evaluates desktop navigation capabilities using visual input and pointer control, GPT-5.4 attained a 75% score. Human testers averaged 72.4% on this benchmark.
OpenAI characterizes this model as its most accurate to date. Factual errors decreased by 33% compared to GPT-5.2’s baseline.
Access and Cost Structure
GPT-5.4 deployment begins today for ChatGPT Plus, Team, and Pro subscribers under the designation GPT-5.4 Thinking. The GPT-5.2 Thinking variant remains operational through June 5, 2026, providing a three-month transition period.
Enterprise and Edu account administrators can activate early access features through their control panels. GPT-5.4 Pro is provisioned for Pro and Enterprise subscription levels.
API access costs $2.50 per million input tokens and $15 per million output tokens. These rates exceed GPT-5.2 pricing, which stood at $1.75 and $14 respectively.
Context window support extends to one million tokens within Codex. Standard API implementations accommodate 272,000 tokens, with expanded requests incurring double billing rates.
According to OpenAI, GPT-5.4’s improved token efficiency compensates for elevated per-token costs, potentially lowering overall expenditure.
The model integrates into OpenAI’s Codex development environment and is accessible via standard API as gpt-5.4.
