Short answer: “03” most likely means OpenAI o3—but o3 is deprecated
If you searched for the ChatGPT 03 model, you most likely mean OpenAI o3, not a model literally named “03.” OpenAI says o3 is no longer available in ChatGPT and is deprecated in the API. Its API model page still lists technical details and prices, but for a new production integration, start by comparing models currently offered in the API catalog rather than making o3 your default.
ChatGPT and the API are separate products. A model’s former availability or tools in ChatGPT do not establish that the same experience is available through the API. Check the documentation for the specific product surface you plan to use.
What o3 was designed to do
OpenAI describes o3 as a reasoning model for complex tasks. Its API page lists text and image input and text output; it does not list audio or video input for o3. These specifications describe the API listing, not every capability that may have been available in ChatGPT.
That context may help when maintaining an existing integration or comparing its behavior with another model. It does not make o3 the automatic choice for a new system: the model page labels it deprecated. Verify any required API feature in current documentation rather than assuming it is supported because it was available in a ChatGPT experience.
Should you choose o3 for a production API?
For a new production workload, generally do not choose a deprecated model as your starting point. First identify a currently supported candidate in OpenAI’s API catalog, then compare it with o3 only if you have a specific reason, such as an existing dependency, a migration, or test results showing it fits your task. The supplied model-page information does not establish a shutdown date; deprecation is a reason to plan cautiously, not proof that requests have already stopped.
Before building or renewing around o3, verify:
- Availability and retirement timing: confirm the model is enabled for your API account and review the official deprecation information before relying on it.
- Exact model name: the API page lists
o3and the dated snapshoto3-2025-04-16, and marks o3 deprecated. Confirm the identifier and status in current documentation before use. - Input and output fit: the listing specifies text and image input and text output. Check that the precise inputs and API features your application requires are documented for the API.
- Operational limits: check the limits that apply to your account and test the latency and failure behavior your service can tolerate. Do not assume limits are identical across accounts.
- Migration cost: measure how much application logic, evaluation, and prompt adjustment would be needed to move to a currently supported model.
For a current comparison candidate, use the official API model catalog rather than assuming an older announcement’s suggested successor is still the newest or best fit. Choose based on your workload’s quality, latency, and cost requirements.
o3 API pricing listed by OpenAI
The o3 API model page lists standard prices of $2 per million input tokens and $8 per million output tokens. It also lists cached input at $0.50 per million tokens. These are API token prices, not a ChatGPT subscription price. Because the page marks o3 deprecated, treat the listed rates as a reference to that API listing—not a guarantee of continued access or future pricing. Confirm current price and account eligibility before budgeting.
Illustrative aggregate cost calculation: across multiple requests, suppose billed usage totals 1 million input tokens and 250,000 output tokens. At the listed standard rates, the token charges would be $4: 1 × $2 for input + 0.25 × $8 for output. This illustrates token charges only; it does not imply that those quantities fit in a single request or guarantee the total cost of a workload.
How to test whether o3—or another model—is right for your task
Broad capability descriptions do not tell you how a model will perform in your production workflow. Run a controlled comparison on the actual task:
- Collect representative cases. Include routine requests, difficult edge cases, and inputs likely to cause errors.
- Set pass criteria before testing. Define what counts as a correct answer, acceptable structured response, or successful code change. Include human review where errors could create safety, financial, or operational risk.
- Verify required API features. Check current API documentation for the features your system needs, then test those exact flows. Do not assume a ChatGPT feature is available through the API in the same way—or that o3 supports a feature unless its API documentation says so.
- Measure quality, latency, and cost together. Track task success and error types alongside response time and token use. A model that performs better on a difficult case may still miss your latency or budget target.
- Test the migration path. Compare a currently supported candidate with your existing o3 behavior and review failures before switching traffic. Keep a fallback and rollback plan while validating the change.
Choose o3 only if it remains available to your account, passes your workload-specific tests, and the benefits justify relying on a deprecated model. Otherwise, select a supported model from the current API catalog and validate it against the same cases.