Core Insights - OpenAI has introduced the GPT-5.4 model, which features native computer control capabilities, allowing it to operate software, browse the web, and execute tasks using mouse and keyboard controls, significantly enhancing its integration with enterprise applications like Excel and Google Sheets [2][3][4] Group 1: Model Capabilities - GPT-5.4 is designed for complex knowledge work and enterprise applications, marking a shift from being merely a conversational tool to an automated task execution system [4] - The model has achieved a success rate of 75.0% in the OSWorld-Verified desktop navigation benchmark, surpassing the human average of 72.4% and significantly improving from the previous version's 47.3% [4][8] - In financial services, GPT-5.4's performance in internal benchmarks has increased from 43.7% to 88.0%, demonstrating its enhanced capabilities in financial modeling tasks [4][18] Group 2: Tool Search Mechanism - The introduction of a "Tool Search" mechanism in GPT-5.4 reduces token consumption by allowing the model to retrieve tool definitions only when needed, rather than preloading all definitions, leading to a 47% reduction in total token usage during tasks [12][13] Group 3: Financial and Enterprise Integration - The launch of the "OpenAI Financial Services" suite includes ChatGPT for Excel and Google Sheets, enabling direct integration into spreadsheets for complex financial modeling and analysis [18] - In internal investment banking benchmarks, GPT-5.4 achieved an average score of 87.3% in modeling tasks, significantly higher than the previous version's 68.4% [18] Group 4: Performance Metrics - In the GDPval test, GPT-5.4 reached or exceeded the performance of industry professionals in 83.0% of cases, an increase from 71.0% in the previous version [19] - The model has shown a 33% reduction in factual error rates compared to GPT-5.2, indicating improved accuracy in outputs [22] Group 5: Pricing and Efficiency - The pricing for GPT-5.4 has increased compared to GPT-5.2, with input costs at $2.50 per million tokens and output costs at $15 per million tokens, reflecting its enhanced capabilities [26][28] - OpenAI justifies the price increase by highlighting the model's superior performance in complex tasks and the efficiency of its new reasoning mechanisms, which reduce token consumption [28]