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

# claude-haiku-4-5

> Anthropic's small-and-cheap model — for classification, routing, structured extraction at scale. Vision and tool use included; pair with prompt caching on the system prompt for the lowest possible per-request cost.

`claude-haiku-4-5` is the small / fast / cheap Claude. It's the right default for **anything you'd run at scale** where the prompt is short, the response is short, and you'd rather call the API ten thousand times an hour than three hundred. Vision and tool use are still included; what you give up is the bigger models' reasoning depth on hard problems.

**Pricing:** $1.00 / 1M input, $5.00 / 1M output, \$0.10 / 1M cache read — see the [rate card](/pricing).

## Protocols

| Protocol                | Path                                                |
| ----------------------- | --------------------------------------------------- |
| Anthropic Messages      | `POST https://llm.bytespike.ai/v1/messages`         |
| OpenAI Chat Completions | `POST https://llm.bytespike.ai/v1/chat/completions` |

## Quickstart

<CodeGroup>
  ```bash cURL theme={null}
  curl https://llm.bytespike.ai/v1/messages \
    -H "x-api-key: $BYTESPIKE_API_KEY" \
    -H "anthropic-version: 2023-06-01" \
    -H "content-type: application/json" \
    -d '{
      "model": "claude-haiku-4-5",
      "max_tokens": 256,
      "messages": [
        { "role": "user", "content": "Classify: refund / billing / technical / other.\n\nMy invoice charged twice." }
      ]
    }'
  ```

  ```python Python (anthropic SDK with cache_control) theme={null}
  import anthropic

  client = anthropic.Anthropic(
      base_url="https://llm.bytespike.ai",
      api_key=os.environ["BYTESPIKE_API_KEY"],
  )

  resp = client.messages.create(
      model="claude-haiku-4-5",
      max_tokens=256,
      system=[
          {
              "type": "text",
              "text": LARGE_CLASSIFICATION_RUBRIC,
              "cache_control": {"type": "ephemeral"},
          }
      ],
      messages=[{"role": "user", "content": "My invoice charged twice."}],
  )

  print(resp.content[0].text)
  ```
</CodeGroup>

## Capabilities

| Capability                      | Supported   |
| ------------------------------- | ----------- |
| Chat completions                | ✅           |
| Streaming (SSE)                 | ✅           |
| Vision (image input)            | ✅           |
| Tool use (function calling)     | ✅ parallel  |
| Prompt caching (cache\_control) | ✅           |
| Extended thinking               | —           |
| Web search                      | —           |
| JSON / structured output        | ✅           |
| Context window                  | 200K tokens |

## When to use

* **Classification, routing, triage.** Tickets → categories. Emails → priority. Calls → next-best action.
* **Structured extraction at scale.** PII redaction, entity extraction, schema-driven parsing on a firehose.
* **Cache-amortized agents.** Big system prompt + many short user turns; cache\_control on the system makes each turn \~10× cheaper after the first.
* **Vision OCR on cheap models.** Haiku's vision is good enough for receipts, invoices, screenshots — at a quarter of Sonnet's price.

When **not** to use:

* Hard reasoning — no extended thinking on Haiku; go to Sonnet or Opus.
* Long-form writing — Haiku's prose tier is below Sonnet.

## Next

* [claude-sonnet-4-6](/models/claude-sonnet-4-6) — production mid-tier
* [claude-opus-4-8](/models/claude-opus-4-8) — 200K-context flagship