Key Developments

OpenAI today released GPT-5.4 Thinking, marking a significant leap in AI capabilities with native computer-use functionality. The model is now available to ChatGPT Plus, Team, and Pro users, replacing the previous GPT-5.2 version.

The standout feature is GPT-5.4’s ability to directly interact with websites and software systems - making it OpenAI’s first general-purpose model with built-in computer control. The model also delivers 33% fewer factual errors compared to its predecessor and achieves 89.3% accuracy on web browsing tasks, setting a new industry benchmark.

Meanwhile, Anthropic introduced their Automated Alignment Agent (A3) framework for safety mitigation, and Intel released OpenVINO 2026.0 with expanded LLM support for their CPU, NPU, and GPU products.

Industry Context

This release represents a critical shift toward agentic AI - systems that can perform real-world tasks autonomously rather than just responding to queries. The computer-use capabilities position GPT-5.4 as a direct competitor to Anthropic’s Claude Computer Use, intensifying the race for AI that can operate independently across digital environments.

The enhanced factual accuracy addresses one of the most persistent criticisms of large language models, while improved tool calling makes the system more reliable for developers building automated workflows.

Practical Implications

For Irish and European businesses, GPT-5.4’s capabilities could accelerate automation initiatives, but adoption may face regulatory headwinds. Recent PwC research shows only 9% of Irish companies have widespread AI agent deployment compared to 52% in the US, with many citing data integration challenges.

Developers can now build more sophisticated agents that navigate complex multi-step processes across different applications. However, the computer-use features raise important questions about security, privacy, and compliance with the EU’s AI Act, which some experts warn may limit European experimentation with high-risk AI applications.

Open Questions

Key uncertainties remain around pricing for the enhanced capabilities, specific limitations of the computer-use functionality, and how these features will interact with European data protection requirements. The gap between US and EU AI adoption rates may widen further as American companies gain access to more advanced autonomous capabilities.

Source: OpenAI