Tool support requires the appropriate API integration; a supported tool is not automatically active in every chat.
Supported API features and tools
Streaming responses
Function calling and structured outputs
Image input and prompt caching
Web search, hosted shell, and skills
Before you choose
Deprecated: OpenAI schedules API shutdown for April 1, 2027.
No native audio or video support.
Fine-tuning is not supported.
A coding model does not automatically have repository access or permission to run commands.
Choose the reasoning effort
lowmediumhighxhigh
Choose the effort explicitly when your integration exposes it. Compare a moderate setting with higher effort on your own engineering tasks. No default is marked because the reviewed model reference does not establish one. Require actual test output before treating a patch as verified.
Unsupported settings: none, minimal, max.
Use the Responses API. Chat Completions and Batch are not supported by this model’s endpoint table.
272,000 tokens is the documented maximum input; the 400,000-token context window includes room for output.
The listed model alias is gpt-5.3-codex. Tools must be configured by the integration; do not assume another model’s tool list applies.
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
Diagnose a failing regression
Ground the investigation in one failing test and a narrow code path.
Investigate this failing test using the supplied test output and relevant source files. State the expected behavior, trace the smallest relevant execution path, and identify the likely cause with file and function references. Propose a minimal patch and a regression test. Do not change unrelated code, suppress the assertion, or claim a test passed unless you actually ran it and can report the result.
Workflow 02
Implement a bounded refactor
Make preserved behavior and forbidden changes explicit before editing.
Refactor the duplicated validation in these functions into one shared implementation. Preserve public interfaces, error messages, and documented edge cases. First list the behavior that must remain unchanged. Then provide a minimal diff and tests covering each caller. Avoid unrelated formatting or dependency changes. If execution tools are unavailable, label the tests as proposed rather than executed.
Workflow 03
Evaluate a model migration
Build a repeatable evaluation before replacing an existing coding integration.
Design a migration evaluation for our current coding assistant using the tasks and constraints below. Include representative bug fixes, feature changes, and negative cases where no edit is needed. Define pass criteria for correctness, patch scope, test honesty, latency, and cost. Specify the evidence to capture and a rollback rule. Do not assume the replacement has the same API defaults or tool support.
Developer reference
OpenAI API pricing
These are OpenAI API reference prices, not EZ Ai Assist subscription prices.
These are Standard token rates. No separate long-context rate is listed for this model in the reviewed pricing table.
Cached-input pricing applies only to eligible cached tokens. Tools and other service tiers can have separate charges.
A published price does not extend the model’s lifetime. Plan migration before April 1, 2027 and check current documentation before use.
Common questions
A few things worth knowing.
Is GPT-5.3 Codex being retired?
Yes. OpenAI announced its API deprecation on October 1, 2026, with shutdown scheduled for April 1, 2027. OpenAI recommends GPT-6 Sol. This is an API lifecycle notice, not a statement that EZ Ai Assist has already removed the model.
Does this guide give the model access to my repository?
No. Repository files, shell execution, and tool permissions must be supplied by the surrounding application. In a plain chat, paste only relevant authorized excerpts and ask for a proposed patch.
Can I use Chat Completions for this model?
The model reference lists Responses as supported and Chat Completions and Batch as unsupported. Check endpoint support when migrating; do not assume every OpenAI coding model uses the same interface.
Why are context and maximum input different?
OpenAI lists a 400,000-token context window, a 272,000-token maximum input, and up to 128,000 output tokens. The context number is not permission to send 400,000 input tokens.
Which tools are documented?
The model’s Responses tool list includes function calling, web search, hosted shell, and skills. Each needs an appropriate integration. Do not infer file search or computer use support from a different model’s page.
Are these my subscription charges?
No. These figures describe direct OpenAI API token usage. EZ Ai Assist subscriptions, access, and limits are separate; check the pricing page and your app workspace.
How can I check whether a patch is safe?
Review the actual diff, run the relevant tests, and inspect failure paths. Separate code the model proposed from commands it executed. Keep a rollback path and evaluate the replacement on the same cases before migration.
Check the source
Official documentation
Specifications and API prices checked on . Example prompts and workflow advice are editorial guidance from EZ Ai Assist.