OpenAI GPT-5.5 Pro (Text + Vision)

Overview

GPT-5.5 Pro is OpenAI’s pro variant of GPT-5.5. It uses extra test-time compute to push accuracy on hard tasks.

You give it written instructions and, if needed, images for visual context. It returns a text answer that can include step-by-step plans, code, or structured data when you ask for it. It’s useful when correctness matters more than speed.

It supports a 1,050,000-token context window and up to 128,000 output tokens. Its knowledge cutoff is December 1, 2025.

What you can build

• Code review and refactor assistants for large repositories
• Long-form technical writing with citations and cross-checking via web search
• Data analysis helpers that explain results in plain language
• Visual QA apps that answer questions about screenshots, charts, and UI states
• Contract and policy analysis that extracts obligations, risks, and edge cases
• Research workflows that gather sources, reconcile conflicts, and write summaries

Inputs

• A required written prompt that describes the task, question, or desired output.
• Optional image files to provide visual context (up to 10 images per request). Use PNG, JPG/JPEG, WEBP, or non-animated GIF.
• Optional system-level instructions that define rules, tone, or formatting (up to 5,000 characters).
• Optional identifiers for the end user and the conversation session. Use a string or number to keep chat history consistent.
• An optional reasoning effort setting to control how hard the model thinks on the request (medium, high, or xhigh).
• An optional verbosity setting to control response length (low, medium, or high).
• An optional web-search switch that lets the model browse the web during the task when it needs fresh information.

Outputs

• A text response with the final answer, such as an explanation, plan, report, or code.
• When you ask for structured output, the response can be formatted as JSON-like structured data inside the text.
• If you enable tool use (like web search), the response can include the results woven into the final write-up.

Limitations

• Output is text only. It does not output images, audio, or video.
• Image input is supported, but only as input. Audio and video inputs are not supported.
• Streaming responses are not supported.
• Some advanced tool categories are not available on this model (for example, computer-use style UI control).
• Very hard requests can run for a long time. Break big tasks into smaller steps when possible.
• Low-quality inputs reduce reliability. Blurry screenshots, tiny text, scanned pages, heavy compression, or cluttered images can cause errors.
• The model’s built-in knowledge stops at December 1, 2025. Use web search for anything that changed after that date.

Safety & compliance

OpenAI applies safety policies and monitoring to reduce harmful outputs. The GPT-5.5 system card describes expanded safeguards for high-risk domains, including cybersecurity and biological topics.

Don’t request disallowed content such as instructions for wrongdoing, sexual content involving minors, or targeted harassment. Avoid sending sensitive personal data unless you have a legal basis and user consent.

If you provide images, only send content you have rights to share. Don’t upload NSFW images, watermarked assets, or content with embedded logos when it isn’t necessary for the task.