Claude Haiku 5.5 API
anthropic/claude-haiku-5.5Claude Haiku 5.5 turns tickets, documents, and screenshots into labels, structured fields, and short summaries, with a 1M-token context and 128K-token replies. Set output_config.effort per request to keep simple routing fast and give harder extractions more reasoning.
Claude Haiku 5.5
Anthropic · messages
Quick start
Send a POST request to /v1/messages with anthropic/claude-haiku-5.5. Choose JSON or SSE with stream, and receive generated content and usage.
- Endpoint
- POST /v1/messages
- Model ID
- anthropic/claude-haiku-5.5
- Protocol
- Claude Messages
Request fields
| Field | Requirement | Description |
|---|---|---|
| model | Required | anthropic/claude-haiku-5.5 |
| messages | Required | A non-empty ordered array with role and content; content accepts text or image blocks. |
| max_tokens | Required | Output token limit from 1 to 128,000; the playground defaults to 4,096. |
| system | Optional | System instructions as a string or text blocks. |
| output_config.format | Optional | Use json_schema and schema to define structured text output. |
| thinking | Optional | Defaults to adaptive thinking. |
| cache_control | Optional | Use an ephemeral cache marker with a 5m or 1h ttl; the minimum prompt length is 512 tokens. |
| tools | Optional | Define callable tools with name, description, and input_schema. |
| tool_choice | Optional | Use {"type":"auto"} for automatic tool selection or {"type":"none"} for a text reply. |
| stop_sequences | Optional | Strings that stop generation. |
| metadata | Optional | Additional request metadata such as user_id. |
| stream | Optional | true returns SSE events; false returns complete JSON. The API defaults to false; the playground uses true. |
Response and usage
Successful responses include generated content and usage. Billing settles from actual input, output, and cache tokens.
Streaming
Set stream: true to read SSE. Finalize usage accounting from the terminal usage event.
curl --request POST \
--url "https://api.vidgo.ai/v1/messages" \
--header "x-api-key: $VIDGO_API_KEY" \
--header 'anthropic-version: 2023-06-01' \
--header 'Content-Type: application/json' \
--data '{
"model": "anthropic/claude-haiku-5.5",
"messages": [
{
"role": "user",
"content": "Review this function and suggest a fix for the edge case."
}
],
"max_tokens": 4096,
"stream": false
}'Store API keys on your server and call the model from server-side code.
Continue with
Related Models
Claude Haiku 5.5 API FAQ
What is the Claude Haiku 5.5 API?
Claude Haiku 5.5 is an Anthropic model that turns text and images into labels, extracted fields, summaries, and short answers, and it is the fastest model in the Claude 5.5 lineup. It is built for high-volume work such as routing, classification, and subagent tasks. Adaptive thinking follows the effort setting, within a 1M-token context and up to 128K output tokens. Use the API tab for Claude Messages and Chat Completions examples, or try a conversation in the Playground.
What changed in Claude Haiku 5.5 from Haiku 4.5?
Context grows from 200K to 1M tokens and output from 64K to 128K tokens. Reasoning is set with output_config.effort instead of budget_tokens, and requests omit temperature, top_p, top_k, and assistant prefill. Anthropic reports 39.2% on Terminal-Bench 4.0, up from 0.0% for Haiku 4.5.
How do I set Claude Haiku 5.5 effort?
In Claude Messages, set output_config.effort to low, medium, high, xhigh, or max; medium is the default. Low effort keeps routing and tagging fast, while high or above helps with long or ambiguous documents. Thinking is on by default and can be disabled at high effort or below.
Can Claude Haiku 5.5 extract fields into JSON?
Yes. Define the fields in output_config.format with type json_schema, then send the document text or an image of it. The reply follows the schema, so invoices, contracts, and forms can feed databases without manual parsing.
What does Claude Haiku 5.5 cost on Vidgo?
Input costs $0.08 (16 credits) and output costs $0.40 (80 credits) per 1M tokens. Cache reads cost $0.008, five-minute cache writes $0.10, and one-hour cache writes $0.16 per 1M tokens. Anthropic lists $0.10 input and $0.50 output for prompts up to 100K tokens.
Is Claude Haiku 5.5 good for coding subagents?
It works well for scoped lookups, file inspections, and summaries that a lead agent hands off. Anthropic reports 39.2% on Terminal-Bench 4.0 and recommends Claude Sonnet 5.5 or Opus 5.5 for complex agentic coding. A common setup pairs a Sonnet or Opus lead with several Haiku 5.5 workers.
Can Claude Haiku 5.5 read screenshots and charts?
Yes. Claude Messages requests accept image blocks together with text, so you can ask about a chart, a form, or an interface screenshot. The model replies in text, for example with extracted values or a short explanation.