o3 API pricing
Standard tier, blended at 3:1 input to output: $3.50 per million tokens.
OpenAI charges $2.00 per million input tokens and $8.00 per million output tokens for o3. It is sold in 5 forms, from $1.75 to $35.00 blended; the table below shows what each costs against standard. 6 of the models nearest it in price are cheaper on a blended basis.
Every tier of o3
Blended at 3:1 input to output. The last column compares each against the standard tier.
| Source | ||||||||
|---|---|---|---|---|---|---|---|---|
| Standard Standard | o3 | OpenAI | $2.00 | $8.00 | $0.500 | $3.50 | Standard | ↗ |
| Standard | o3 | OpenRouter | $2.00 | $8.00 | $0.500 | $3.50 | Same | ↗ |
| Dated snapshot | o3-2025-04-16 | OpenAI | $2.00 | $8.00 | $0.500 | $3.50 | Same | ↗ |
| Batch | o3:batch | OpenRouter | $1.00 | $4.00 | $0.250 | $1.75 | −50% | ↗ |
| Pro | o3-pro | OpenRouter | $20.00 | $80.00 | Not listed | $35.00 | 10× | ↗ |
Prices last checked: . Methodology
Which tier to use
What each costs against standard in our data, and the job it is for.
Batch
The batch tier (o3:batch) is 50% cheaper than standard. Batch requests are queued and answered asynchronously, typically within 24 hours, so it suits work nobody is waiting on: evaluations, back-fills, bulk classification and tagging. Anything a user is waiting on belongs on the standard tier.
Pro
o3-pro is 10× standard. Pro tiers spend more compute on each request. At 10 times the price, route only the hard cases to it, the ones where o3 gets the answer wrong, and keep everything else on standard.
Dated snapshots
One dated snapshot (o3-2025-04-16) is listed at the same price as standard. A snapshot pins the model to one release, while the undated name moves to newer releases when the vendor updates it. Pin a snapshot in production when your prompts depend on exact output; use the undated name to pick up improvements automatically.
Direct or through a reseller
Through OpenRouter, o3 is the same price as buying direct. A reseller is worth it for one bill across vendors and automatic fallback; buy direct for the vendor's own batch, caching and rate-limit terms.
Against the models nearest in price
Standard tiers, blended at a 3:1 input to output ratio.
| Source | ||||||
|---|---|---|---|---|---|---|
| o3 This model | OpenAI | $2.00 | $8.00 | 200,000 | $3.50 | ↗ |
| GPT-4.1 | OpenAI | $2.00 | $8.00 | 1,047,576 | $3.50 | ↗ |
| Llama 3.1 405B Instruct | Together AI | $3.50 | $3.50 | 4,096 | $3.50 | ↗ |
| GPT-5.1 | OpenAI | $1.25 | $10.00 | 272,000 | $3.44 | ↗ |
| GPT-5 | OpenAI | $1.25 | $10.00 | 272,000 | $3.44 | ↗ |
| Gemini 2.5 Pro | Google AI | $1.25 | $10.00 | 1,048,576 | $3.44 | ↗ |
| GPT-5.1 Codex | OpenRouter | $1.25 | $10.00 | 400,000 | $3.44 | ↗ |
| Mistral Large | OpenRouter | $2.00 | $6.00 | 128,000 | $3.00 | ↗ |
| Grok 4.5 | xAI | $2.00 | $6.00 | 500,000 | $3.00 | ↗ |
Prices last checked: . Methodology
Questions
How much does o3 cost per million tokens?
$2.00 per million input tokens and $8.00 per million output tokens, with cached input at $0.500. At a 3:1 input to output mix that blends to $3.50. Checked 28 Oct 2026.
How much cheaper is o3 through the Batch API?
o3:batch costs $1.00 in and $4.00 out per million tokens, 50% cheaper than standard, in exchange for asynchronous delivery.
What does o3 Pro cost?
o3-pro costs $20.00 in and $80.00 out per million tokens, 10× standard.
What would o3 cost for a real workload?
A thousand requests a day at 2,000 input and 600 output tokens each comes to $264.00 a month on the standard tier, or $132.00 through batch.
What is the context window for o3?
200,000 tokens of input, with up to 100,000 tokens of output.
GPT-4.1 · Llama 3.1 405B Instruct · GPT-5.1 · GPT-5 · Gemini 2.5 Pro · All OpenAI models · Cost calculator