# o3-mini

The cost-efficient member of the o3 reasoning family, providing strong chain-of-thought performance on STEM and coding tasks at a price point that makes per-request reasoning practical.

- **Model ID:** `openai/o3-mini`
- **Type:** chat
- **Providers:** azure
- **Context window:** 200,000
- **Maximum output tokens:** 100,000
- **Pricing:** $1.1/1M input tokens, $4.4/1M output tokens
- **Canonical page:** https://vercel.com/ai-gateway/models/o3-mini

## Supported parameters

Detailed capability metadata has not been reported for this model.

## Example

```ts
import { streamText } from 'ai'

const result = streamText({
  model: 'openai/o3-mini',
  prompt: 'Why is the sky blue?'
})
```

## About

o3-mini was released on January 31, 2025 as the cost-efficient tier of the o3 reasoning model family. It continues the pattern established by o1-mini: delivering strong chain-of-thought reasoning on structured domains (mathematics, coding, science) at a fraction of the full model's cost.

The model supports the `reasoning_effort` parameter, letting you control reasoning depth per request. Low effort for straightforward technical queries conserves tokens and reduces cost; high effort for competition-level problems applies the full reasoning capability. This flexibility lets you use o3-mini as the default for all technical queries rather than maintaining a routing layer.

With a context window of 200K tokens and support for the standard API features, o3-mini handles the same types of requests as full o3. The tradeoff is concentrated in reasoning depth: on the hardest problems, full o3 will produce more thorough analysis.

## What to consider

o3-mini makes chain-of-thought reasoning affordable enough to run on every request rather than reserving it for the hardest problems. The `reasoning_effort` parameter enables further cost optimization.

Like o1-mini before it, o3-mini concentrates its reasoning capability on structured problem domains rather than broad general knowledge.

## When to use

### Best For

- **Math and science reasoning:** Competition-level problems, derivations, and quantitative analysis at accessible cost
- **Code reasoning:** Algorithm analysis, debugging, and optimization with step-by-step deliberation
- **High-frequency reasoning pipelines:** Per-request chain-of-thought on technical workloads at scale
- **Education platforms:** Tutoring and problem-solving assistance with visible reasoning steps
- **Cost-optimized reasoning:** Tasks that benefit from deliberation but don't justify full o3 pricing

### Consider Alternatives When

- **Maximum reasoning quality:** Full o3 for the hardest problems where every increment of accuracy matters
- **Broader knowledge needed:** Full o3 or GPT-5 for tasks requiring wide-ranging factual recall
- **Fastest reasoning:** o4-mini for a newer cost-efficient reasoning option with vision support
- **General-purpose tasks:** GPT-5 mini for workloads that don't benefit from chain-of-thought

## Best for

- **Math and science reasoning:** Competition-level problems, derivations, and quantitative analysis at accessible cost
- **Code reasoning:** Algorithm analysis, debugging, and optimization with step-by-step deliberation
- **High-frequency reasoning pipelines:** Per-request chain-of-thought on technical workloads at scale
- **Education platforms:** Tutoring and problem-solving assistance with visible reasoning steps
- **Cost-optimized reasoning:** Tasks that benefit from deliberation but don't justify full o3 pricing

## Consider alternatives

- **Maximum reasoning quality:** Full o3 for the hardest problems where every increment of accuracy matters
- **Broader knowledge needed:** Full o3 or GPT-5 for tasks requiring wide-ranging factual recall
- **Fastest reasoning:** o4-mini for a newer cost-efficient reasoning option with vision support
- **General-purpose tasks:** GPT-5 mini for workloads that don't benefit from chain-of-thought

## Frequently asked questions

### How does o3-mini compare to o1-mini?

o3-mini is the next generation of cost-efficient reasoning, delivering stronger performance on key benchmarks while maintaining the affordability that makes per-request reasoning practical.

### Does o3-mini support the `reasoning_effort` parameter?

Yes. You can control reasoning depth per request, enabling cost optimization across mixed-difficulty workloads.

### What context window does o3-mini support?

200K tokens, matching the o3 family.

### When should I use full o3 instead?

When the hardest problems require maximum reasoning depth and the quality gap between mini and full is consequential for your application.

### How does AI Gateway handle authentication for o3-mini?

AI Gateway accepts a single API key or OIDC token for all requests. You don't embed OpenAI credentials in your application; AI Gateway routes and authenticates on your behalf.

### What are typical latency characteristics?

This page shows live throughput and time-to-first-token metrics measured across real AI Gateway traffic.

## Links

- [Model page](https://vercel.com/ai-gateway/models/o3-mini)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://platform.openai.com/docs/models/o3-mini)
- [Provider pricing](https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service)