OpenAI launches GPT-5.4 with Pro and Thinking versions
Summary
OpenAI has released GPT-5.4, a new foundation model positioned as its most capable and efficient frontier model for professional work. It is available in a standard version, a reasoning model (GPT-5.4 Thinking), and a high-performance variant (GPT-5.4 Pro). The API version features context windows up to 1 million tokens and improved token efficiency. GPT-5.4 achieved record scores on OSWorld-Verified and WebArena Verified computer use benchmarks, an 83% on OpenAI’s GDPval test for knowledge work, and leads Mercor’s APEX-Agents benchmark for professional skills. The model also demonstrates a 33% reduction in individual claim errors and an 18% overall error reduction compared to GPT 5.2. A new Tool Search system for the API optimizes tool calling by allowing on-demand definition lookups, and a new safety evaluation assesses chain-of-thought integrity.
Key takeaway
For CTOs and VPs of Engineering evaluating new AI models for enterprise deployment, GPT-5.4 presents a compelling option due to its significantly improved performance benchmarks, reduced error rates, and enhanced API features like the 1 million token context window and Tool Search. Consider piloting GPT-5.4 for applications requiring extensive context, high accuracy in knowledge work, or complex multi-tool integrations to capitalize on its efficiency and capability gains.
Key insights
GPT-5.4 offers enhanced capabilities, efficiency, and safety features for professional AI applications.
Principles
- Larger context windows improve model utility.
- Token efficiency reduces operational costs.
Method
The Tool Search system allows models to dynamically look up tool definitions, reducing token consumption and speeding up requests in multi-tool environments.
In practice
- Utilize GPT-5.4 Pro for high-performance tasks.
- Employ GPT-5.4 Thinking for complex reasoning.
- Integrate Tool Search for efficient API tool calling.
Topics
- GPT-5.4
- Large Language Models
- AI Benchmarks
- Context Windows
- AI Safety
Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, Machine Learning Engineer, AI Product Manager
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by AI News & Artificial Intelligence | TechCrunch.