OpenAI's latest flagship model processes one million tokens of context while controlling computers directly, arriving in enterprise data clouds through a $200 million Snowflake partnership announced earlier this year.
GPT‑5.4 introduces agentic AI capabilities that allow it to perform computer actions like clicking a mouse and editing files autonomously, marking OpenAI's first general-purpose model that can actually operate PCs rather than just provide instructions.
The model achieves 75% accuracy on operating system benchmarks and supports smooth integration into existing workflows without requiring custom designs. A massive context window expansion enables processing of up to one million tokens in a single input, a increase from the 272,000 tokens supported by GPT‑5.3. This capacity allows professionals working on extensive coding projects, legal document analysis, or multi-step problem-solving to tackle intricate workflows without breaking down or simplifying inputs.
Dynamic tool loading reduces token usage by 47%, optimizing resource allocation for tasks like debugging and data analysis. Interruptible reasoning allows users to modify the model's approach during a task without restarting sessions, providing greater control over complex operations.
Performance benchmarks show a 20% improvement over GPT‑5.2 with enhanced coding capabilities, faster outputs, and improved UI design precision. The model is available through ChatGPT, OpenAI API, and OpenAI's coding tool Codex.
Snowflake and OpenAI announced a multi-year $200 million strategic partnership on February 2 that brings advanced AI models directly into Snowflake's enterprise platform. Under the deal, OpenAI models including GPT‑5.2 will be natively integrated into Snowflake's Cortex AI services and Snowflake Intelligence, making them accessible to more than 12,600 customers worldwide across all major cloud platforms.
The integration allows organizations to build and deploy context-aware AI applications and agents on their proprietary data without moving it outside secure environments. Developers can call OpenAI models directly using standard SQL commands while keeping sensitive corporate information within Snowflake's perimeter.
Graphic design platform Canva plans to use the partnership to expand its visual AI offerings while keeping user data strictly governed. WHOOP uses the integration to provide deeper, context-aware performance analytics for athletes through embedded models that strengthen both reasoning capability and data governance.
This follows a similar $200 million deal Snowflake signed with Anthropic in late 2025 to bring Claude models into its platform, signaling a multi-model strategy where Snowflake acts as the neutral, secure layer for all top-tier AI providers.
