Short answer
OpenAI o3 is a reasoning model intended for complex, multi-step tasks involving text, code, and images. As of November 9, 2026, OpenAI lists o3 as deprecated and says it will be removed from the API on December 11, 2026. OpenAI identifies GPT-5 as its successor. That makes o3 a model to evaluate for an existing integration or migration—not a sensible default for a new production system. Check the o3 model page and API deprecations guidance before planning a change.
What “reasoning model” means for choosing o3
OpenAI describes o3 as a model for multi-step problems that involve analysis across text, code, and images. That positioning makes it a candidate to assess when a task requires connected steps—for example, interpreting a chart image and explaining its implications. It does not guarantee a correct answer or establish that o3 is the right fit without testing your own workload.
For a production decision, measure the outputs your application actually receives against your accuracy, latency, cost, and safety requirements. A model’s reasoning label is not a substitute for those checks.
Capabilities and API details OpenAI lists
OpenAI’s model documentation lists text input and output, image input, and text output. It specifies a 200,000-token context window and a maximum of 100,000 output tokens. The page also lists streaming, function calling, and Structured Outputs as supported. These are documented model capabilities, not a guarantee that every request or workflow will meet your needs. Confirm the specific endpoint and feature behavior for your implementation in the o3 API documentation.
For example, a team could evaluate image input by providing representative charts and scoring whether the model extracts the relevant values correctly. A coding team could test whether it identifies a defect in a representative change and whether its proposed fix passes the team’s tests. These are evaluation ideas, not reported o3 test results.
Treat outputs as model-generated rather than verified facts. For consequential work, define review or validation appropriate to the impact of an error. The model page does not establish that o3 is reliable enough for any particular deployment.
Should you choose o3 for a production workload?
For a new integration, shortlist models currently offered in the API rather than building around deprecated o3. OpenAI says o3 is succeeded by GPT-5, and has announced that o3 will be removed from the API on December 11, 2026. If you already use o3, treat migration planning as an operational task and test a replacement before switching traffic.
A practical decision rule:
- New production work: shortlist currently available models from the API catalog and test them on representative tasks. Do not make o3 the default just because it was designed for reasoning-heavy work.
- Existing o3 deployment: find every use of the
o3model ID, plan for the announced shutdown, and compare a replacement before moving traffic. - Tasks involving images or multi-step analysis: verify that the candidate model and endpoint support the specific inputs and tool behavior you need, then test on representative and difficult cases.
- Strict cost or latency limits: compare measured end-to-end results for your prompts and workflow. The reasoning label alone does not establish your application’s speed or cost.
For an apples-to-apples evaluation, run representative inputs through o3, if your organization can still access it, and candidate replacements. Score task success, critical errors, latency, and total cost under the same conditions. Include cases where the model should ask for clarification or avoid taking an action. This is a proposed evaluation method, not a claim that one model wins.
Production API status, endpoints, limits, and pricing
The official model page lists both the alias o3 and the dated snapshot o3-2025-04-16 as deprecated. It lists two supported inference endpoints: Responses (POST /v1/responses) and Chat Completions (POST /v1/chat/completions). The model page also identifies image input and text output. Endpoint support does not mean that your organization still has access or that the model will remain available through the announced shutdown.
OpenAI’s API reference gives December 11, 2026 as the announced shutdown date for o3 and o3-2025-04-16. To monitor model shutdowns programmatically, OpenAI’s deprecations guidance says to check the optional shutdown_date field through the List models endpoint (GET /v1/models) or Retrieve a model endpoint. Check your organization’s access and applicable rate limits separately; they can depend on your account and usage tier.
The o3 model page lists standard API token prices of $2.00 per 1 million input tokens, $0.50 per 1 million cached input tokens, and $8.00 per 1 million output tokens. These are API token prices, not ChatGPT subscription prices. The model page displays a separate lower-rate table as well; because its pricing-mode label is not clear in the retrieved text, this guide does not use those figures for an estimate. Check the current pricing display and your account’s availability before budgeting.
Illustrative calculation: 1 million input tokens plus 100,000 output tokens at the listed standard rates would be $2.00 + $0.80 = $2.80. This is arithmetic based on the displayed standard rates, not a quote or assurance that your organization can use o3.
Before deploying or migrating, confirm:
- Whether your organization can still access the model and is prepared for the December 11, 2026 shutdown.
- Whether your integration uses
o3oro3-2025-04-16, and calls a supported endpoint. - Whether the image input, function calling, structured output, or streaming behavior required by your workflow is supported in that integration.
- The rate limits and pricing that apply to your account and request pattern.
- That a replacement meets your task-quality and operational acceptance criteria before moving production traffic.
Bottom line
OpenAI documents o3 as a reasoning model with image input and support for the Responses and Chat Completions endpoints. Its deprecated status and announced December 11, 2026 API shutdown are central to the selection decision: assess currently available models for new work, and verify access, plan migration, and test a replacement if you already depend on o3.