claude-haiku-5-5
Claude Haiku 5.5 is an Anthropic model for high-volume, latency-sensitive tasks such as classification, extraction, and routing. It accepts text and image inputs, produces text, and supports adaptive thinking, a 1M-token context window, and up to 128K output tokens.
Modalities
In / Out Price
10% off$0.09/ 1M
$0.1 Official price
$0.45/ 1M
$0.5 Official price
Context
1M
Released
Oct 6, 2026
API call examples
POST /v1/chat/completionscurl --request POST 'https://api.tokatlas.com/v1/chat/completions' \
--header "Authorization: Bearer $UNIROUTER_API_KEY" \
--header 'Content-Type: application/json' \
--data '{
"model": "claude-haiku-5-5",
"messages": [
{
"role": "user",
"content": "Explain quantum entanglement."
}
],
"temperature": 0.7
}'Providers & routes
Current provider · Anthropic
Platform configured price · Group: ccmax
USD / 1M tokens. Each row shows input and output from the same tier. Displayed rates use the lowest available group multiplier; your API key group may differ.
Claude Haiku 5.5 for frequent classification and extraction
A candidate for classification, extraction, and lightweight document review. Define a fixed output shape and an explicit value for uncertain cases.
High-volume steps
Start with summaries, classification, and extraction into a fixed format.
Text and images
Compare screenshots with descriptions and return findings supported by the inputs.
Longer inputs
Supply documents alongside definitions and keep passages available for comparison.
Adaptive reasoning
Test rare categories at different effort levels and identify cases for human review.
About
High-volume, bounded steps
Haiku 5.5 fits extraction, labeling, support triage, and small subagent tasks. Process well-formed items first and escalate unclear or contradictory results for deeper review.
Haiku 4.5, Haiku 5.5, and Sonnet
The longer window makes room for policy documents, label definitions, and the working data together. Include rare labels and long inputs in tests, then compare batch accuracy and review costs when deciding where to use Sonnet.
Haiku 4.5 to 5.5 migration differences
Migration and long inputs
Update the model ID and check the default adaptive effort. Long prompts may cross pricing thresholds. Measure context size and retries in cost estimates, and use the pricing section for actual rates.
Separate routing from drafting in support. Extract issue type, relevant policy, and urgency before application rules assign the case. Use real request lengths and review samples to assess whether long prompts change batch economics.
claude-haiku-5-5 FAQ
Which coding steps are a good fit for Haiku 5.5?
Try it on file discovery, issue triage and contained edits. For changes that span modules, compare acceptance results with a model such as Sonnet.
Is it useful in a high-volume workflow?
Make extraction, summarization and routing separate steps with clear acceptance checks, then track time and success rate for each step.
How should I size the output for a long input?
The published standard limits are one million context tokens and 128K output tokens. Thinking uses the output budget too, so constrain the fields and length of an extraction response.
When is Sonnet 5.5 worth trying instead?
If the task spans several sources, involves repeated tradeoffs or regularly requires manual fixes, compare Sonnet on the same inputs and track completion rate.