Tool support requires the appropriate API integration; a supported tool is not automatically active in every chat.
Supported API features and tools
Streaming responses and function calling
Text and image understanding
Web search, file search, and tool search
Image generation through a tool
Apply patch and computer use
MCP integrations
Before you choose
Structured outputs are not supported.
Code interpreter and hosted shell are not supported.
No native audio or video support; image generation requires a tool.
Fine-tuning is not supported.
Some requests can take several minutes; more compute does not remove the need for review.
Choose the reasoning effort
medium · defaulthighxhigh
Begin with the documented medium default and a clearly bounded review. Compare high or xhigh only on tasks where they produce a measurable improvement. Keep response time, cost, and reviewer effort in your evaluation rather than optimizing a single score.
Unsupported settings: none, minimal, low, max.
Use the Responses API. Chat Completions and Batch are not supported by this model’s endpoint table.
For requests that may take minutes, use background mode to reduce timeout risk.
The documented snapshot is gpt-5.4-pro-2026-03-05. Function calling support does not imply structured-output support.
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
Test a difficult technical argument
Look for unsupported steps and counterexamples in supplied reasoning.
Review this technical argument and its stated assumptions. Identify steps that do not follow from the evidence, hidden dependencies, and plausible counterexamples. Cite the relevant passage for each issue and distinguish a confirmed error from an unresolved question. Suggest a targeted test or revised assumption for every important gap. Do not claim a proof or experiment was verified unless you can show the supporting evidence.
Workflow 02
Assess a risky system change
Turn a proposed change into a focused risk and validation review.
Assess this proposed system change using the architecture notes, incident history, and rollout constraints provided. Identify failure modes involving compatibility, partial rollout, retries, and recovery. For each, state the evidence, likely impact, detection signal, and mitigation. Separate supported risks from hypotheses. Finish with a go/no-go checklist and rollback criteria, without changing the proposed scope.
Workflow 03
Compare competing explanations
Ask what evidence would distinguish hypotheses, not which story sounds best.
Given these observations and three proposed explanations, compare what each explanation predicts, which supplied facts support or contradict it, and what remains unknown. Do not invent measurements or treat correlation as proof. Rank the next tests by how well they distinguish the explanations and by effort. Provide a provisional conclusion only if the current evidence supports one.
Developer reference
OpenAI API pricing
These are OpenAI API reference prices, not EZ Ai Assist subscription prices.
There is no cached-input discount listed for GPT-5.4 Pro; the table contains only input and output rates.
Above 272,000 input tokens, the 2× input and 1.5× output rates apply to the full session, not just the excess tokens.
This table covers Standard processing. Regional processing adds 10% where available; tool usage can add charges. Pricing-table entries alone do not establish endpoint support.
Confirm the current model reference, processing tier, and prices before integrating.
Common questions
A few things worth knowing.
How is GPT-5.4 Pro different from GPT-5.4?
Pro is intended for harder problems with more compute and has medium, high, and xhigh efforts. Its endpoint and feature restrictions are different: it is not a drop-in replacement for every standard GPT-5.4 integration.
Can GPT-5.4 Pro return schema-constrained structured outputs?
The model reference lists structured outputs as unsupported, while function calling is supported. Do not equate those two features. Validate any application-specific output before using it.
Can it run Python or a hosted shell?
Code interpreter and hosted shell are not supported for this model in the reviewed reference. It can propose code in text, but that is not evidence that the code was executed.
Can it stream, and why might I use background mode?
Streaming is supported. Some difficult requests can still take several minutes, so OpenAI recommends background mode to reduce timeout risk for long-running work.
What are the context-related costs?
The context window is 1,050,000 tokens, but higher rates start above 272,000 input tokens and apply to the full session. There is no cached-input discount listed. Estimate cost before sending a very large task.
Does Pro mean my EZ Ai Assist Pro subscription?
No. This model name and its direct API rates are separate from an EZ Ai Assist subscription. Check the app for current model availability and the website’s pricing page for plan details.
When should I compare another model?
Compare alternatives when you need unsupported tools, lower cost, or shorter turnaround. Use a small set of representative tasks with explicit success criteria and review the output quality before switching a production workflow.
Check the source
Official documentation
Specifications and API prices checked on . Example prompts and workflow advice are editorial guidance from EZ Ai Assist.