Microsoft has announced the general availability of OpenAI’s GPT-5.4 in Microsoft Foundry, introducing a new generation of AI models designed specifically for reliable production deployments and agent-driven workflows. Unlike earlier models that primarily focused on intelligence and experimentation, GPT-5.4 is built to support real-world enterprise applications where consistency, reliability, and long-running workflows are critical.
As organizations increasingly deploy AI agents across business processes, one of the biggest challenges has been maintaining context, stability, and execution accuracy across multi-step tasks. GPT-5.4 addresses these challenges by combining enhanced reasoning capabilities with dependable task execution, allowing AI systems to not only plan work but also complete complex tasks across tools, data sources, and workflows with minimal manual intervention.
One of the key advancements in GPT-5.4 is its improved reasoning consistency across long interactions. The model is designed to maintain intent and context over extended multi-turn conversations, reducing task drift and improving reliability when AI is used for longer processes such as research, document analysis, and automation workflows. This makes it particularly valuable for organizations deploying AI-powered agents for customer support, operational automation, and enterprise knowledge work.
GPT-5.4 also introduces stronger instruction alignment, which helps reduce the amount of prompt engineering required to guide the model’s responses. By better adhering to instructions and maintaining structured outputs, the model allows teams to deploy AI solutions that require less oversight and tuning. Additionally, improvements in latency and performance enable faster responses for real-time workflows, making the model more practical for interactive applications and operational systems.
Another notable capability is the integration of computer-use functionality, which allows GPT-5.4 to interact with files, tools, and applications in a structured way. This enables more sophisticated automation scenarios, including data extraction, workflow orchestration, guarded code execution, and coordinated agent handoffs. These capabilities allow AI agents to perform tasks across multiple systems while maintaining reliability and security.
For organizations requiring deeper analysis, Microsoft also introduced GPT-5.4 Pro, a premium version of the model designed for scenarios where analytical depth is more important than speed. GPT-5.4 Pro supports multi-path reasoning evaluation, allowing the model to explore multiple potential solutions before selecting the most appropriate one. This makes it particularly useful for scientific research, financial analysis, and complex decision-making workflows where thorough reasoning and accuracy are essential.
Both GPT-5.4 and GPT-5.4 Pro are available through Microsoft Foundry, Microsoft’s enterprise platform for building and managing AI applications. Foundry provides organizations with the governance and operational controls required for responsible AI deployment, including policy enforcement, monitoring, version management, and auditability. These capabilities allow enterprises to integrate advanced AI systems into their existing environments while maintaining compliance and security.
The new model supports large context windows of up to 272K tokens and beyond, enabling it to process extensive documents, datasets, and workflows within a single interaction. GPT-5.4 is available through Standard Global and Standard Data Zone (US) deployments, with additional regional availability expected soon.
Overall, the introduction of GPT-5.4 in Microsoft Foundry represents a significant step forward for organizations looking to move from AI experimentation to reliable, production-grade AI systems. By combining stronger reasoning, structured tool usage, and enterprise governance capabilities, GPT-5.4 enables businesses to build intelligent agents and automation solutions that can execute complex workflows at scale with greater consistency and reliability.