o3 is a deep reasoning model in OpenAI's o series, suited for tasks where answers are not intuitive and require integrating multiple conditions. It focuses on programming, mathematics, scientific problems, and visual analysis, incorporating information from images into text-based reasoning. For work that requires checking assumptions, comparing approaches, and explaining conclusions, o3 offers more value than simply pursuing brief, instant responses.
Clarify capacity, input and output, and invocation methods before selecting a model.
Visual-text reasoning and custom function calling are model capabilities, while session management and parameter settings are invocation endpoint features; each parameter must be applicable to o3 to take effect.
Learn what o3 can bring to your work.
o3 is suited to problems involving multiple conditions that cannot be addressed by directly applying a template. When handling mathematics, science, or solution analysis, you can ask it to state assumptions clearly, compare candidate explanations, and provide checkable reasoning. Its value lies in integrated analysis, rather than compressing complex problems into an unsupported conclusion.
o3 does more than describe image content; it can also reason about chart trends, whiteboard relationships, and conditions in diagrams. Submitting an image together with a specific question, such as asking it to explain an unusual change or identify contradictions between conditions, makes better use of its visual analysis capabilities than simply asking “what is in the image?”
Through custom function calls, o3 can participate in multi-step workflows involving queries, calculations, and result interpretation. The application provides tool definitions, executes requests, and returns results, after which the model continues its analysis based on them. This makes it possible to combine reasoning with real data, while tool permissions and execution environments remain managed by the application.
Start with specific tasks to find where the model can be effective.
Provide the relevant code, error logs, expected behavior, and reproduction conditions, and let o3 compare possible causes and propose fixes and testing recommendations. Deliverables can include issue identification, change explanations, and validation steps, making it suitable for debugging that requires understanding multiple constraints rather than merely completing a short piece of code.
Submit business charts, whiteboard photos, or process diagrams, along with review objectives, and let o3 explain key relationships, distinguish observations from assumptions, and organize questions that need verification. It is suitable for creating review outlines and solution comparison notes; important data should ideally also be provided as text to facilitate checking the image-reading results.
Provide the problem, known conditions, and existing approaches, and let o3 check whether the reasoning omits conditions, try different approaches, and propose hypotheses that can be further verified. It can be used to organize proof drafts, research questions, and experimental discussion outlines, though final conclusions should still be tested through independent calculations or professional methods.
Choose based on task complexity, input materials, and expected results.
When a task requires deeper integrated judgment, especially when involving code, scientific conditions, or cross-analysis of text and images, prioritize o3. o4-mini, on the other hand, focuses on fast, cost-efficient reasoning and is better suited to large volumes of repetitive tasks. For actual model selection, use the same set of business examples to compare answer quality and processing efficiency, rather than relying only on model names.
o3-pro is an independent variant with extended thinking and an emphasis on answer reliability, not a request switch for o3. When using o3, existing messages integrations can use Chat Completions; new workflows organized around responses and tools can use Responses; if you want to ask follow-up questions continuously through a conversation ID, you can use the AI Chat entry point.
From a small-scale task to formal integration.
Define the goal, required inputs, and output requirements, using real business examples as a starting point.
Open the trial page, confirm the parameters supported by this entry point, then submit a small-scale task to view the results.
Keep the complete model ID, use the request format specified in the documentation, and confirm billing rules on the Pricing page.
Before formal use, understand output quality and capability scope.
Answers to common questions about using o3.
No. o3, o3-mini, and o3-pro are different public models. When calling o3, use model: "o3". o3 is designed for complex reasoning and visual analysis; do not directly apply the parameters or capabilities of other models to it just because their names are similar.
In Chat Completions messages, submit the text question together with an image_url content block, and use model: "o3". The question should ideally target a specific relationship or judgment objective; if the image contains key numerical values, you can also provide a text version to reduce reading errors.
Generating code and executing code are two different things. o3 can analyze code and request calls to defined functions, but execution must be completed by the application or tool environment. To verify the effect of a fix, run tests and feed the results back so the model can continue making judgments based on actual results.
Responses provides the reasoning setting, while Chat Completions provides the reasoning_effort setting. When using them, choose a configuration accepted by o3; do not directly apply tiers from other models. You can also explicitly ask in the prompt to check assumptions, list verification steps, and control response length.
When using Chat Completions, you typically submit the required history through messages; when using the AI Chat conversation entry point, you can enable stateful and include the same id in subsequent requests. Regardless of the method, when key conditions change, state them clearly to avoid relying on old assumptions.
Model information · Updated: 2026-10-01. For calling parameters and billing rules, see the API and pricing sections.
Start with a clear goal and evaluate whether it fits your work based on real results.
Input
$0.40per 1M tokens
Official $2.00
Output
$1.60per 1M tokens
Official $8.00
You pay$13,584for 142,800 credits · $0.0951 / credits46% OFF
You're billed in credits. Pick a top-up tier to see its per-credit rate — the larger the top-up, the cheaper each credit.
This is only a basic call example. See the full docs for more parameters and advanced usage.
View Docs