openai / gpt-5.4-pro

## Overview GPT 5.4 Pro is OpenAI’s high-compute GPT 5.4 variant for hard problems. It “thinks harder” before it answers, which improves multi-step reasoning and code accuracy. You can give it text plus images, and it returns text that explains, decides, or drafts deliverables. It’s useful when you need fewer retries on complex work and you want very large context for long sessions. ## What you can build - Visual Q&A for screenshots, UI bug reports, and step-by-step troubleshooting - Document-heavy analysis workflows, like contract issue spotting and policy comparisons - Long-context code review and refactor planning across many files - Data extraction from charts, tables, and dense diagrams in images - Agent-style task plans that break big work into verifiable steps - High-precision writing: specs, technical docs, and structured business memos ## Inputs - A required instruction prompt written in plain text. - Optional image attachments for the model to interpret (PNG, JPEG/JPG, WEBP, or non-animated GIF). - If you attach multiple images, keep them focused and relevant to the same task. - Optional system-level guidance that sets rules, style, and constraints for the whole response. - Optional reasoning effort selection with 3 levels: medium, high, or extra-high (xhigh). - Optional verbosity control to request shorter or longer answers (low, medium, or high). - Optional web search toggle, if you want the model to look up fresh information during the run. - Optional user and session identifiers, if you want the app to keep chat history across turns. ## Outputs - A text response that follows your instructions, with explanations, steps, or final deliverables. - If you provide images, the response can include grounded descriptions of what the model sees. - If web search is enabled in the run, the response may incorporate newly found facts from the web. ## Limitations - Output is text only. It does not generate images, audio, or video. - Audio and video inputs are not supported for this model. - GPT 5.4 Pro supports only medium, high, and extra-high reasoning effort. It won’t accept “none”. - The model has a large context window (up to 1,050,000 tokens) and can generate up to 128,000 tokens. - The knowledge cutoff for GPT 5.4 Pro is August 31, 2025. It may miss newer events. - For high-resolution visual understanding, OpenAI documents higher-fidelity image processing modes. The highest fidelity supports up to 10.24M total pixels or a 6000-pixel maximum dimension. - Low-quality inputs can break results. Expect errors with blurry screenshots, tiny text, heavy compression, or skewed scans. - Like all LLMs, it can hallucinate. Verify critical claims, citations, numbers, and code behavior. ## Safety & compliance - Use the model in ways that follow OpenAI’s Usage Policies and Service Terms. - Don’t submit sensitive personal data unless you have a clear legal basis and user consent. - The model may refuse requests for disallowed content, including harmful wrongdoing guidance. - If you enable web search, review results before acting. Web content can be incorrect or malicious.