# Claude Haiku 5.5 - Core.Today AI API

> Anthropic's newest Haiku model (October 2026), the fast, low-cost choice for high-volume work such as classification, routing, extraction, and sub-agent tasks. $0.10/$0.50 per M tokens for prompts up to 100K tokens; prompts over 100K tokens (cache reads and writes included) bill the whole request at $0.50/$2.50. 1M-token context window, 128K max output tokens, adaptive thinking on by default, prompt caching (read $0.01/M, 5-minute write $0.125/M) and web search.

- **Provider**: Anthropic
- **Model ID**: claude-haiku-5-5
- **Category**: LLM
- **Credits**: 1 per 1K tokens (avg)
- **Speed**: Fast
- **Quality**: High

## Model Specifications
- **Context Window**: 1M tokens
- **Max Output**: 128K tokens
- **Training Cutoff**: Not published
- **Supported Formats**: text, json, markdown
- **Compatible SDK**: Anthropic, OpenAI

### Capabilities
- Vision (image input)
- Function Calling
- Streaming
- JSON Mode
- System Prompt

### Token Pricing (per 1M tokens)
- **Input Tokens**: 185.8 credits ($0.12)
- **Output Tokens**: 929 credits ($0.62)
- **Cached Tokens**: 18.58 credits ($0.01)

## Features
- Lowest-cost current Claude model
- 1M token context window
- 128K max output tokens
- Adaptive thinking on by default (effort default: medium)
- Prompt caching (read 10% / 5-min write 125% of input price)
- Long-context tier: prompts over 100K tokens bill the whole request at 5x
- Web search tool support
- Vision and tool use
- Streaming support

## Use Cases
- High-volume classification and routing
- Structured extraction
- Sub-agent and worker tasks in agent pipelines
- Latency-sensitive chat
- LLM-as-judge at scale

## API Endpoint

Base URL: https://api.core.today
Endpoint: POST /llm/anthropic/v1/messages

## Authentication
Header: Authorization: Bearer YOUR_API_KEY

Note: LLM endpoints use OpenAI-compatible format with Authorization Bearer token.

## Input Parameters

### Required
- **messages**: array - Array of message objects (OpenAI format)

### Optional
- **max_tokens**: integer (default: 4096) - Maximum tokens in response, thinking included (up to 128000)
- **stream**: boolean (default: false) - Enable Server-Sent Events streaming

## Examples

### Ticket Routing
Classify support tickets into queues at Haiku-tier cost
```json
{
  "model": "claude-haiku-5-5",
  "messages": [
    {
      "role": "system",
      "content": "Classify the ticket into exactly one of: billing, account, bug, feature_request, other. Reply with the label only."
    },
    {
      "role": "user",
      "content": "I was charged twice for my October subscription. Can you refund one of them?"
    }
  ],
  "max_tokens": 2000
}
```

### Structured Extraction
Pull fields out of free-form text as JSON
```json
{
  "model": "claude-haiku-5-5",
  "messages": [
    {
      "role": "system",
      "content": "Extract company, amount, currency and due_date from the invoice text. Return JSON only."
    },
    {
      "role": "user",
      "content": "Invoice from Acme Corp. Total due: 1,250,000 KRW by November 30, 2026."
    }
  ],
  "max_tokens": 2000
}
```

## Response Format
```json
{
  "id": "chatcmpl-abc123",
  "object": "chat.completion",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Response text here"
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 100,
    "completion_tokens": 50,
    "total_tokens": 150
  }
}
```

## Tips
- Keep prompts at or under 100K tokens (cache reads and writes count) — above that the whole request, output included, bills at 5x
- Thinking is on by default and counts against max_tokens; set effort to low for simple classification
- Prompt caching: read $0.01/M (10%), 5-minute write $0.125/M — cache large repeated contexts
- Web search costs about 18.6 credits per request ($10 per 1,000 searches) on top of token usage
- Non-default `temperature`, `top_p`, `top_k`, `budget_tokens` and assistant prefill return 400 — omit them (the console strips `temperature` automatically)
- US-only inference (inference_geo: us) bills token charges at 1.1x, including the over-100K tier
- Batch API is not available through the gateway

## Documentation
https://docs.anthropic.com/en/docs/about-claude/models