o1-pro
o1-pro is a reasoning-focused chat model from OpenAI in the o1 series, positioned as a variant that spends additional inference compute on each response.
API Pricing
Cheapest on OpenAI — 40% below avg| Provider | Input / 1M | Output / 1M |
|---|---|---|
| $75.00 | $300.00 | |
| $150.00 | $600.00 | |
| $150.00 | $600.00 |
Prices updated daily. Last check: Aug 31, 2026
o1-pro pricing by provider
Input, output, and batch rates, plus alternatives, for one provider at a time.
Performance & Benchmarks
Source: Artificial Analysis →Benchmarks measured Aug 2026. Scores are independent evaluations, not vendor-reported.
Model Details
General
- Creator
- OpenAI
- Modalities
- Text
Capabilities
- Tool Calling
- No
- Open Source
- No
Strengths & Limitations
Strengths
- Part of OpenAI's o1 reasoning series, which performs internal reasoning before emitting a final answer
- Higher inference-time compute allocation than standard o1, aimed at accuracy on hard multi-step problems
- Suited to problems where a single correct answer is worth more than a fast answer, such as math, science, and complex debugging
- Reduces the need for user-side prompt scaffolding like manual chain-of-thought or self-consistency sampling
- Fits verification and second-opinion roles behind cheaper models in a tiered pipeline
- Available through OpenAI's API surface, so it can slot into existing OpenAI client integrations
Limitations
- Reasoning models add latency; our benchmark record contains no measured time-to-first-token or output throughput to size expectations from
- Output billing includes hidden reasoning tokens, making per-request cost harder to predict than for standard chat models
- Proprietary — weights are not available for self-hosting, so provider choice is limited
- We do not track a confirmed context window, modality support, or tool-calling support for this model, so those must be verified against provider documentation
- The high-compute configuration is generally a poor fit for latency-sensitive or high-volume traffic
- Superseded in OpenAI's own lineup by later reasoning generations, so availability on third-party endpoints may be limited
Key Features
About o1-pro
Common Use Cases
o1-pro targets workloads where correctness per attempt outweighs speed and cost per attempt: competition-style math, quantitative and scientific derivations, algorithm design, root-cause analysis in large codebases, and structured reasoning over dense technical documents. It also works well as the escalation step in a tiered architecture — route the bulk of traffic to a fast general-purpose model, then send low-confidence or high-stakes cases to o1-pro for a more deliberate answer or an independent verification pass. It is a weak fit for chat interfaces with tight latency budgets, bulk classification, summarization at scale, or any high-volume pipeline where the added reasoning tokens dominate the bill.
Frequently Asked Questions
How much does o1-pro cost?
Pricing varies by provider and by pricing type (input tokens, output tokens, cached input, and any batch or committed-use discounts), and it changes over time. See the pricing table on this page for the current figures. Note that for reasoning models like o1-pro, internal reasoning tokens are counted as output, so real-world cost per request is usually higher than the visible answer length suggests.
What is o1-pro best used for?
Hard, multi-step problems where accuracy matters more than latency — mathematical and scientific reasoning, complex debugging, algorithmic design, and analytical work that a faster model tends to get wrong on the first pass. It is also useful as a verification or escalation layer behind a cheaper model.
How does o1-pro differ from o1 and o1-mini?
All three are OpenAI reasoning models from the same generation. o1-pro is the higher-compute configuration, allocating more inference-time reasoning per request than standard o1, while o1-mini is the smaller, faster option. The practical trade-off across the three is deliberation depth versus latency and cost.
Why is o1-pro slower than a normal chat model?
It generates internal reasoning tokens before returning a final answer, and the "pro" configuration deliberately spends more compute on that step. That work happens before the visible response, so time-to-first-token and total response time are typically longer than for a standard instruction-tuned chat model. Our benchmark record for o1-pro contains no measured throughput or latency values, so you should measure against your own prompts.
Does o1-pro accept image input or support tool calling?
We do not currently track confirmed modality or tool-calling support for o1-pro in our database, so we can't state it either way here. Check the documentation of the provider you plan to use — reasoning-model endpoints sometimes expose a different feature set than standard chat endpoints.