> ## Documentation Index
> Fetch the complete documentation index at: https://docs.routeway.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Models & Providers

> How Routeway routes requests across providers and how to pick the right model.

Routeway gives you access to 200+ models from providers like OpenAI, Anthropic, Google, Mistral, DeepSeek, xAI, and more — all through a single API key and a single base URL.

## How Routing Works

When you send a request, Routeway inspects the `model` field, resolves the correct upstream provider, forwards your request using that provider's credentials, and returns a normalized response in the standard OpenAI format.

```mermaid theme={null}
flowchart LR
    A[Your App] -->|model: claude-opus-4-5| B[Routeway API]
    B --> C{Route}
    C -->|OpenAI models| D[OpenAI]
    C -->|Claude models| E[Anthropic]
    C -->|Gemini models| F[Google]
    C -->|Other models| G[...]
    D & E & F & G -->|Normalized response| B
    B --> A
```

You never manage provider credentials, SDKs, or per-provider quirks. Routeway handles it.

***

## Model IDs

Use the provider-native model ID as the `model` value. Routeway maps it automatically.

```python theme={null}
# OpenAI model
client.chat.completions.create(model="gpt-4o", ...)

# Anthropic model — same client, same endpoint
client.chat.completions.create(model="claude-opus-4-5", ...)

# Google model
client.chat.completions.create(model="gemini-2.5-pro", ...)

# DeepSeek model
client.chat.completions.create(model="deepseek-chat", ...)
```

The full list of supported model IDs is on the [Models](/getting-started/models) page.

***

## Choosing a Model

No single model is best for every task. Use this table as a starting point:

| Use case                      | Recommended models                            | Why                                |
| ----------------------------- | --------------------------------------------- | ---------------------------------- |
| General chat & Q\&A           | `gpt-4o-mini`, `gemini-2.0-flash`             | Fast, cheap, high quality          |
| Complex reasoning             | `o3`, `o4-mini`, `claude-opus-4-5`            | Extended thinking, better accuracy |
| Code generation               | `gpt-4o`, `kimi-k2.7-code`, `claude-opus-4-5` | Strong code benchmarks             |
| Long documents                | `gemini-2.5-pro`, `claude-opus-4-5`           | 1M+ token context windows          |
| Image understanding           | `gpt-4o`, `gemini-2.5-pro`                    | Best vision capabilities           |
| Low latency / high throughput | `gpt-4o-mini`, `gemini-2.0-flash`             | Sub-second TTFT at scale           |
| Cost-sensitive pipelines      | `deepseek-chat`, `gpt-4o-mini`                | Lowest cost per token              |

<Tip>
  When evaluating models, run the same set of representative prompts through several candidates and compare quality, latency, and cost. The best model is the one that meets your quality bar at the lowest cost.
</Tip>

***

## Model Capabilities at a Glance

<CardGroup cols={2}>
  <Card title="Text generation" icon="text">
    All models. The core capability — generating coherent, contextual text from a prompt.
  </Card>

  <Card title="Vision / multimodal" icon="image">
    `gpt-4o`, `gpt-4o-mini`, `claude-opus-4-5`, `gemini-2.5-pro`. Send images and PDFs alongside text.
  </Card>

  <Card title="Tool calling" icon="wrench">
    `gpt-4o`, `gpt-4o-mini`, `claude-*`, `gemini-*` and most frontier models. Let models invoke your functions.
  </Card>

  <Card title="Structured outputs" icon="braces">
    `gpt-4o`, `gpt-4o-mini`. Guarantees JSON schema conformance.
  </Card>

  <Card title="Reasoning" icon="brain">
    `o3`, `o4-mini`, `claude-opus-4-5` (extended thinking), `deepseek-r1`. Extended deliberation before answering.
  </Card>

  <Card title="Prompt caching" icon="database">
    `gpt-4o`, `claude-*`, `gemini-*`. Reuse processed prefix tokens to cut cost and latency.
  </Card>
</CardGroup>

***

## Switching Models

Because Routeway uses the OpenAI-compatible format, switching models is a one-line change:

```python theme={null}
# Before
response = client.chat.completions.create(model="gpt-4o", messages=messages)

# After — same client, same code, different model
response = client.chat.completions.create(model="claude-opus-4-5", messages=messages)
```

<Warning>
  Some parameters are model-specific. For example, `reasoning_effort` applies to OpenAI `o`-series models, while Anthropic's extended thinking uses a `thinking` block. Routeway normalizes responses but does not translate provider-specific request parameters.
</Warning>

***

## Model Versioning

Frontier models are regularly updated. To avoid unexpected behavior changes in production, pin to a specific model snapshot:

```python theme={null}
# Pinned — behavior is stable
model = "gpt-4o-2024-11-20"

# Floating alias — always the latest, may change
model = "gpt-4o"
```

Use floating aliases during development to get the latest improvements, then pin to a snapshot before deploying to production.

***

## Pricing

Each model has its own input and output token pricing. Costs are deducted from your Routeway balance per request.

* **Input tokens** — tokens in your `messages` array (including the system prompt)
* **Output tokens** — tokens generated by the model
* **Reasoning tokens** — internal thinking tokens on reasoning models (billed separately)
* **Cached tokens** — prompt tokens served from cache (discounted on supported models)

See the [Models](/getting-started/models) page for per-model pricing and the [Billing](/getting-started/billing) page for balance and usage details.