claude-haiku-5-5 API
Claude Haiku 5.5 is Anthropic’s small model for high-volume tasks, latency-sensitive interactions and subagents, including classification, routing, extraction and summaries. It supports adaptive thinking; XiuAI cost comparisons should also account for input-length pricing tiers.[1][2][3]
claude-haiku-5-5Model overview
Published provider specifications. Available features depend on the current API route.
- API routes
- OpenAI-compatible API / openai-response / openai-response-compact / Anthropic Messages / Gemini GenerateContent / openai-alpha-search
High-volume tasks with clear boundaries
Evaluate Haiku for separate tasks such as classification, request routing, field extraction and short summaries. Define allowed outputs, failure handling and escalation conditions, and compare end-to-end completion and cost rather than response speed alone.[1]
A subagent still relies on the application for context, tool permissions and result integration. Compare the same task with Sonnet or Opus when constraints or reasoning become complex. Small-model positioning does not guarantee lower cost for every task.[1][3]
Migration and conversation handling
Haiku 5.5 uses the newer tokenizer, enables adaptive thinking by default and requires omitting temperature, top_p and top_k. Re-measure the same input when comparing older Haiku versions instead of reusing their token estimates or thinking settings.[1]
Anthropic states that its thinking blocks work only in the producing account or a linked account. Multi-account or multi-model orchestration must account for conversation-state boundaries; retaining text is not the same as retaining all reasoning state.[1]
API access
XiuAI access and tiered costs
Use claude-haiku-5-5 with your existing XiuAI key and the protocol shown here. The current catalog distinguishes input-length pricing tiers and separates input, output, cache reads and writes. Check the applicable tier and request record rather than applying the lowest tier to every long request.[3][1]
Send a request using your existing API key.
POST /v1/chat/completionscurl 'https://router-api.xiu.ai/v1/chat/completions' \
-H "Authorization: Bearer $XIUAI_API_KEY" \
-H 'Content-Type: application/json' \
--data-raw '{
"model": "claude-haiku-5-5",
"messages": [
{
"role": "user",
"content": "Hello"
}
]
}'Read the API documentationFrequently asked questions
Why can long requests use different rates on the same model?
The current Haiku 5.5 catalog has input-length tiers. After crossing a tier boundary, the lowest rates are not the basis for all input, output and cache charges. Check the pricing conditions and actual request record.[3][1]
Can I reuse Haiku 4.5 token estimates after switching?
Do not reuse them unchanged. Anthropic documents a newer tokenizer in 5.5, so the same text can have a different count. Recheck input and output usage before estimating the actual effect of model pricing.[1]
Is Haiku suitable for every subagent task?
Validate by task. Clear, bounded work is a candidate, while extended reasoning, complex coding and multi-constraint decisions should be compared with other models. The application should check outputs and handle escalation instead of assigning every subtask solely by unit price.[1][3]
Related models
claude-sonnet-5-5claude-opus-5-5Sources and updates
Specifications reference provider documentation; prices and availability come from the current XiuAI catalog.
Content reviewed: 2026-11-16
[1] Claude Haiku 5.5 model reference[2] Claude Haiku 5.5 announcement[3] Current XiuAI model and pricing dataModel documentationTool callingStructured outputContext documentation