# GPT-4.1 nano

The fastest and most affordable model in the GPT-4.1 family, built for classification, autocomplete, and lightweight reasoning at the highest throughput and lowest cost.

- **Model ID:** `openai/gpt-4.1-nano`
- **Type:** chat
- **Providers:** azure
- **Context window:** 1,047,576
- **Maximum output tokens:** 32,768
- **Pricing:** $0.1/1M input tokens, $0.4/1M output tokens
- **Canonical page:** https://vercel.com/ai-gateway/models/gpt-4.1-nano

## Supported parameters

Detailed capability metadata has not been reported for this model.

## Example

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

const result = streamText({
  model: 'openai/gpt-4.1-nano',
  prompt: 'Why is the sky blue?'
})
```

## About

GPT-4.1 nano was introduced on April 14, 2025 as the smallest and most latency-optimized model in the GPT-4.1 family. OpenAI designed it specifically for tasks where speed and cost efficiency take priority over frontier reasoning depth: classification, autocomplete, routing decisions, and other lightweight inference workloads that need to run at high volume.

Despite being the entry-level tier of the GPT-4.1 family, GPT-4.1 nano posts creditable benchmark scores for its size: 80.1% on MMLU (Massive Multitask Language Understanding) and 50.3% on GPQA (Graduate-Level Google-Proof Q&A). These numbers show that the GPT-4.1 training improvements carried down to the smallest variant. Like its larger siblings, it supports the full context window of 1.0M tokens, which is a notable capability for a model at its price point and enables it to handle tasks that involve reading long inputs even if the outputs remain short.

GPT-4.1 nano inherits the GPT-4.1 family's 75% prompt caching discount and the removal of surcharges for long-context usage. For applications that preload a large knowledge base or system prompt once and then issue many rapid short queries against it, these economics make nano an attractive option for the query stage of a retrieval-augmented pipeline.

## What to consider

For event-driven pipelines that fire many rapid inferences per user action (real-time intent classification, content routing), GPT-4.1 nano's speed and low cost make it practical to run inference inline without queuing.

## When to use

### Best For

- Real-time classification tasks: sentiment analysis, intent detection, and topic labeling at high request volume
- Autocomplete and inline suggestion features requiring sub-second response times
- Routing and triage logic within multi-model pipelines that decide which downstream model handles a request
- Short-answer extraction from long documents, where the context window of 1.0M tokens and nano's low cost combine well
- Cost-sensitive batch jobs where millions of inferences need to run economically

### Consider Alternatives When

- The task requires multi-step reasoning, code generation, or complex instruction following; GPT-4.1 mini or GPT-4.1 provide meaningfully higher capability
- Output quality on nuanced or ambiguous inputs matters significantly; larger models in the family handle edge cases better
- Chain-of-thought reasoning over hard STEM problems is needed; o1-mini or o1 are purpose-built for that use case

## Best for

- **Real-time classification:** Sentiment analysis, intent detection, and topic labeling at high request volume
- **Autocomplete features:** Inline suggestion experiences requiring sub-second response times
- **Routing and triage:** Logic within multi-model pipelines that decides which downstream model handles a request
- **Short-answer extraction:** Pulling answers from long documents where the context window of 1.0M tokens and nano's low cost combine well
- **Cost-sensitive batch jobs:** Millions of inferences that need to run economically

## Consider alternatives

- **Complex reasoning:** GPT-4.1 mini or GPT-4.1 provide meaningfully higher capability for multi-step reasoning, code generation, or complex instruction following
- **Edge-case quality:** Larger models in the family handle nuanced or ambiguous inputs better
- **Hard STEM problems:** o1-mini or o1 are purpose-built for chain-of-thought reasoning on difficult STEM tasks

## Frequently asked questions

### What tasks is GPT-4.1 nano specifically designed for?

OpenAI designed it for classification, autocomplete, and routing where response speed and low cost outweigh the need for frontier reasoning.

### Does GPT-4.1 nano really support a context window of 1.0M tokens?

Yes. All three GPT-4.1 family members share the context window of 1.0M tokens, which is unusual for a model at nano's price and speed tier.

### What benchmark scores does GPT-4.1 nano achieve?

At launch, GPT-4.1 nano scored 80.1% on MMLU and 50.3% on GPQA, showing that the family's training improvements extended to the smallest variant.

### How does GPT-4.1 nano's pricing compare to the rest of the GPT-4.1 family?

See the pricing section on this page for today's rates. AI Gateway exposes each provider's pricing for GPT-4.1 nano.

### Is GPT-4.1 nano suitable as the query model in a RAG pipeline?

Yes. Pairing a large preloaded knowledge base or system prompt (benefiting from the 75% cache discount) with rapid, inexpensive nano queries is a practical pattern for retrieval-augmented generation at scale.

### When should I use nano versus mini versus GPT-4.1?

Nano: classification, routing, autocomplete. Mini: GPT-4o-class quality with lower cost and latency. GPT-4.1: maximum coding and instruction-following accuracy.

### 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/gpt-4.1-nano)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/concepts/models-sold-directly-by-azure?pivots=azure-openai&tabs=global-standard-aoai%2Cstandard-chat-completions%2Cglobal-standard#gpt-5.1)
- [Provider pricing](https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service)


If you encounter incorrect information, can’t find what you need, need a capability Vercel doesn’t currently offer, or have any other feedback to report, please do it by following the instructions at https://vercel.com/feedback.md.