Tool support requires the appropriate API integration; a supported tool is not automatically active in every chat.
Supported API features and tools
Streaming, function calling, and structured outputs
Text/image input and prompt caching
Web search and file search
Code interpreter, hosted shell, and apply patch
Tool search, skills, and MCP
Computer use and image generation through tools
Before you choose
No native audio or video support; image generation requires a tool.
Fine-tuning is not supported.
Tool access requires integration and permission; outputs still need validation.
Choose the reasoning effort
none · defaultlowmediumhighxhigh
The documented default is none. Evaluate an explicit reasoning level for harder analysis rather than assuming it is already enabled. Compare a small representative set at low or medium before paying for higher effort. Keep tasks bounded and measure correctness alongside response time.
Unsupported settings: minimal, max.
Responses, Chat Completions, and Batch are supported. Built-in tools listed here are Responses API tools.
The documented snapshot is gpt-5.4-2026-03-05. Pin a snapshot when repeatable evaluation matters.
API rate limits vary by account tier and have separate long-context limits.
Put it to work
Start with a more useful prompt.
Original examples from EZ Ai Assist. Adapt these to your task and the features available in your workspace.
Workflow 01
Map a multi-document requirement
Find the implementation consequences of requirements spread across documents.
Using these requirements, design notes, and meeting records, build a traceability table from each confirmed requirement to its source, affected component, acceptance test, and unresolved dependency. Flag contradictions and missing decisions. Do not silently reconcile disagreements or invent requirements. End with the smallest implementable first milestone and the evidence needed to approve it.
Workflow 02
Plan a tool-assisted investigation
Separate read-only investigation from actions that change external systems.
Plan an investigation of this operational issue using the tools and permissions I list. For each step, state the question, input, expected evidence, and stop condition. Separate read-only checks from changes that require approval. If a tool or record is unavailable, describe the gap instead of claiming to use it. Finish with a verification checklist and a concise handoff format.
Workflow 03
Review an interface against a specification
Compare visible details with explicit requirements, not personal taste.
Compare the attached interface screenshot with this written specification. List confirmed mismatches with their screen location and relevant requirement. Separate visual observations from behavior that cannot be verified in an image. Prioritize issues by user impact, propose targeted fixes that preserve the existing design, and write an interactive test for every behavior-dependent finding.
Developer reference
OpenAI API pricing
These are OpenAI API reference prices, not EZ Ai Assist subscription prices.
Above 272,000 input tokens, long-context pricing applies to the full session: 2× input and 1.5× output, not only the excess tokens.
This table shows Standard processing. Batch, Flex, and Fast have separate rates; tool usage may add charges.
Regional processing adds 10% where available. Confirm current prices and service-tier eligibility before estimating cost.
Common questions
A few things worth knowing.
How should I compare GPT-5.4 with Mini?
Run both on representative tasks with the same instructions and evidence. Compare correctness, missed requirements, response time, and total cost. Use the larger model when its measured benefit justifies the difference, not simply because the task has a long prompt.
Is reasoning enabled by default?
The API model page lists none as the default. Supported efforts are none, low, medium, high, and xhigh. A surrounding app may choose different settings; this guide does not establish the app’s default.
Can GPT-5.4 operate a computer?
Computer use is listed as a supported Responses tool. It requires the appropriate integration and permissions. A normal text response does not automatically control a browser or desktop.
What happens above 272,000 input tokens?
OpenAI applies higher long-context rates to the full session. The context limit and the billing threshold are different: fitting within the model’s context window does not mean the lower price applies.
Can it return JSON that follows a schema?
Structured outputs are supported. Configure the schema in the API integration and validate the meaning of the values as well as their shape. Ask the model to preserve missing data instead of filling it with guesses.
Does the API table describe my EZ Ai Assist subscription?
No. The table is a developer reference for direct OpenAI usage. EZ Ai Assist plan prices and included features are separate, and tool or model availability should be checked in the app.
How can I make long-document work more reliable?
Identify the exact question, remove irrelevant material, and request source locations for key claims. Spot-check citations and ask for an explicit list of unresolved contradictions. Keep a human review boundary for decisions.
Check the source
Official documentation
Specifications and API prices checked on . Example prompts and workflow advice are editorial guidance from EZ Ai Assist.