GPT-5.4 is OpenAI's general-purpose reasoning model for professional work, combining complex analysis, coding, and visual understanding, with a focus on improving documents, spreadsheets, presentations, and multi-step tool tasks. It is suited for work that requires sustained understanding of materials and producing reviewable results, rather than just brief Q&A. On this platform, you can connect through responsive, message-based, or simplified conversation entry points.
Clarify capacity, input and output, and invocation methods before selecting a model.
Native specifications describe model capabilities; this platform's input organization, tool configuration, and output control are used according to the selected invocation entry point.
Learn what gpt-5.4 can bring to your work.
GPT-5.4 focuses not only on summarizing materials, but also on creating and revising documents, spreadsheets, and presentation content. After providing business context, data definitions, and delivery requirements, it can be used to organize analytical structures, explain calculation logic, draft body text and presentation outlines, and bring content closer to deliverables that can be reviewed and further refined.
It combines the coding capabilities of GPT-5.3-Codex with stronger visual understanding, making it suitable for handling code, errors, and interface screenshots together. Developers can ask the model to propose fixes, generate modified code, and continue iterating based on runtime results; after connecting testing tools, they can also inspect actual page behavior.
GPT-5.4 natively supports computer use and Tool Search, enabling it to determine actions from screenshots and look up tool definitions as needed. For tasks that require reading information, calling business functions, and organizing results, it is suitable as the planning and decision-making core; execution programs complete actions and return results to the model for continued processing.
Start with specific tasks to find where the model can be effective.
Provide operating data text, metric definitions, and chart screenshots, and ask the model to explain changes, distinguish facts from assumptions, and draft an executive summary. Deliverables can include an anomaly list, analysis paragraphs, and a presentation outline; if downloadable spreadsheet or presentation files are needed, connect file-creation tools to complete formatting and export.
Submit requirements, relevant code, error logs, and page screenshots together, and let GPT-5.4 analyze interaction issues, propose component changes, and outline testing steps. It is suitable for tasks involving both code logic and visual layout, delivering fix code, explanations of causes, and an acceptance checklist, then confirming final behavior with real browser tests.
Organize contracts, specifications, or project materials into text with section numbering, and clearly specify the clauses to compare and the review objectives. The model can help organize definitions, identify inconsistencies between earlier and later sections, and create comparison tables and follow-up question lists; retain original-text locations so reviewers can return to the materials for verification rather than receiving only a generic summary.
Choose based on task complexity, input materials, and expected results.
Compared with GPT-5.2, GPT-5.4 places greater emphasis on professional deliverable creation, visual understanding, and multi-step tool collaboration; compared with GPT-5.3-Codex, it incorporates coding capabilities into a more general-purpose work model. If a task includes code, business documents, and interface observation at the same time, GPT-5.4 is worth prioritizing for testing, but this should not be understood to mean it is necessarily stronger for every coding task.
GPT-5.4 is the base model, while GPT-5.4 Pro is a separate variant for more complex tasks, not a configuration switch for the same ID. For integration, complex tool workflows can start with Responses; existing messages-based applications can use Chat Completions; when continuous text conversation is primarily needed, AI Chat can simplify session management.
From a small-scale task to production integration.
Define objectives, required inputs, and output requirements, using real business examples as a starting point.
Open the playground page, confirm the parameters supported by this entry point, then submit a small-scale task to review the results.
Keep the complete model ID, use the request format specified in the documentation, and confirm billing rules on the Pricing page.
Before formal use, understand output quality and capability scope.
Answers to common questions about using gpt-5.4.
GPT-5.4 Thinking is the ChatGPT product form of GPT-5.4; API calls use gpt-5.4. Experiences such as process prompts and interactive adjustments in ChatGPT should not be directly regarded as part of an API response; gpt-5.4-pro is a separate variant.
When using Chat Completions, place text and image_url in the content array of the same message, and explain what you want identified, compared, or interpreted. Images can use auto, low, or high detail; when analyzing small text and complex interfaces, clear screenshots are usually more important than vague questions.
The model has relevant planning and instruction-generation capabilities, but actions require tools provided by the application. After a function call is returned, the program should execute it and send back the result; browser or desktop operations also require the appropriate environment. Sending only a single operation request will not automatically establish a complete execution and verification workflow.
GPT-5.4's official native evaluations used reasoning settings from none to xhigh. On this platform, Responses provides the reasoning object, while Chat Completions provides the reasoning_effort field, but the same set of levels cannot be directly applied to both: the current Chat Completions parameter enumeration does not include none. When configuring, use levels that are valid for gpt-5.4 through the selected endpoint, and compare response quality and wait time with real tasks; values such as minimal and max in shared enumerations should not be directly considered settings supported by this model.
When using a message-based endpoint, continue including relevant conversation history in messages; Responses can organize conversation history in input. To simplify text conversation management, you can choose AI Chat, continuously pass stateful=true and the returned id, and use preset to set the long-term role and response requirements.
Model information · Updated: 2026-10-01. For call parameters and billing rules, see the API and pricing sections.
Start with a clear goal and evaluate whether it suits your work based on real results.
This is only a basic call example. See the full docs for more parameters and advanced usage.
View Docs