OpenAI's most advanced unified model released August 2025, combining reasoning and fast response capabilities in one system. Features adaptive routing that automatically switches between instant answers and deep thinking based on task complexity, achieving 94.6% on AIME 2025 and 74.9% on SWE-bench Verified.
- Unified reasoning system with automatic routing between fast responses and deep thinking modes
- State-of-the-art performance achieving 94.6% AIME 2025, 74.9% SWE-bench Verified, and 84.2% MMMU
- Significantly reduced hallucinations by 45% compared to GPT-4o and 80% reduction in thinking mode
- Enhanced coding capabilities excelling at frontend development and complex multi-step programming tasks
- 400K context window with comprehensive tool integration including web search and image generation
Web Search
Search the web for current info.
JSON Mode
Output responses in valid JSON format.
Image Generator
Generate images from text prompts.
Model Information
Supported Formats
GPT 5.5OpenAI's most advanced model released April 2026. GPT-5.5 is recommended for most reasoning workloads, combining frontier intelligence with configurable reasoning effort (none → xhigh). Available in API, Codex, and ChatGPT.
GPT 5.5 ProHighest-intelligence API option in the GPT-5.5 family. Use for the most challenging problems that require maximum reasoning depth and can tolerate higher latency. Supports full reasoning effort range from none to xhigh.
GPT 5.4GPT-5.4 is OpenAI's frontier model for complex professional work, designed for agentic, coding, and professional workflows with the highest reasoning capability. Features a 1.05M token context window, native computer use, and built-in support for web search, file search, and MCP.