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

# Serverless Inference

> Call any supported LLM via API with pay-per-token pricing and no infrastructure to manage.

Serverless inference lets you call any supported model through a simple API — no GPU provisioning, no deployments, no setup. Pay per token and start immediately.

Navigate to **Deploy → Serverless** from the sidebar to browse models and get your API key.

***

## How It Works

<Steps>
  <Step title="Browse Models">
    View all available serverless models in the model catalog. Each card shows the model name, family, context window, and per-token pricing.
  </Step>

  <Step title="Filter & Select">
    Use the search bar, capability filters (Coding, Reasoning, Agentic, Tool Use, Chat), or use case filters (Production APIs, Automation, Low-Latency, Research, etc.) to find the right model for your workload.
  </Step>

  <Step title="Review Details">
    Select a model to view its full specifications — parameter count, architecture, context window, capabilities, and detailed pricing breakdown.
  </Step>

  <Step title="Call the API">
    Use the provided code examples (cURL, Python, SDK, or CLI) with your API key to start sending requests immediately. No deployment step required.
  </Step>
</Steps>

***

## Pricing

Serverless models use **pay-per-token** pricing, displayed as a rate per 1M tokens on each model card. You are charged for input and output tokens at per-model rates. **Cached tokens are \$0.00** — when Tensormesh serves a token from its KV cache, you are not charged for it.

Current pricing for each model is displayed on the **Deploy → Serverless** page. For a full breakdown, see [Pricing Overview](/pricing-overview).

<Tip>
  Track your token usage and costs at any time under **Operations → Serverless Usage**. See [Serverless Usage](/serverless-usage) for the full per-model breakdown.
</Tip>

***

## Choosing the Right Model

<CardGroup cols={2}>
  <Card title="Reasoning & Agents" icon="brain">
    Look for large MoE models with long context windows. Higher parameter counts generally mean stronger reasoning and tool use for complex agentic workflows.
  </Card>

  <Card title="Coding Agents" icon="code">
    Coding-specialized models with large context windows are best for multi-file tasks, code review, and agentic workflows. Smaller coding models offer a faster, cheaper alternative for simpler tasks.
  </Card>

  <Card title="Low-Latency & High-Throughput" icon="bolt">
    Smaller, compact models are the fastest and most cost-effective — ideal for real-time assistants, chatbots, and high-volume API services where speed matters more than raw capability.
  </Card>

  <Card title="Long-Context & Document Processing" icon="layer-group">
    For workflows that process large documents, codebases, or long conversation histories, prioritize models with the largest context windows available in the catalog.
  </Card>
</CardGroup>

***

## Quick Start

<Tabs>
  <Tab title="cURL">
    ```bash theme={null}
    curl --request POST \
      --url https://serverless.tensormesh.ai/v1/chat/completions \
      --header 'Authorization: Bearer YOUR_API_KEY' \
      --header 'Content-Type: application/json' \
      --data '
    {
      "model": "MODEL_NAME",
      "max_tokens": 16384,
      "temperature": 0.6,
      "messages": [
        {
          "role": "user",
          "content": "Hello, how are you?"
        }
      ],
      "top_p": 1,
      "top_k": 40,
      "presence_penalty": 0,
      "frequency_penalty": 0
    }
    '
    ```
  </Tab>

  <Tab title="Python">
    ```python theme={null}
    import requests

    url = "https://serverless.tensormesh.ai/v1/chat/completions"

    payload = {
        "model": "MODEL_NAME",
        "max_tokens": 16384,
        "temperature": 0.6,
        "messages": [
            {
                "role": "user",
                "content": "Hello, how are you?"
            }
        ],
        "top_p": 1,
        "top_k": 40,
        "presence_penalty": 0,
        "frequency_penalty": 0
    }
    headers = {
        "Authorization": "Bearer YOUR_API_KEY",
        "Content-Type": "application/json"
    }

    response = requests.post(url, json=payload, headers=headers)

    print(response.text)
    ```
  </Tab>

  <Tab title="SDK">
    Install the [Tensormesh Python SDK](/sdk/index) (`pip install tensormesh`), then:

    ```python theme={null}
    from tensormesh import Tensormesh
    from tensormesh.types import ChatMessage

    with Tensormesh(inference_api_key="YOUR_API_KEY") as client:
        completion = client.inference.serverless.chat.completions.create(
            model="MODEL_NAME",
            messages=[
                ChatMessage(role="user", content="Hello, how are you?"),
            ],
        )

    print(completion.choices[0].message.content)
    ```
  </Tab>
</Tabs>

<Info>
  The serverless API is **OpenAI-compatible**. You can use any OpenAI SDK or client library by pointing the base URL to `https://serverless.tensormesh.ai` and using your Tensormesh API key.
</Info>