anthropic

Anthropic: Claude Haiku 4.5

anthropic/claude-haiku-4.5

A Claude family candidate for short interactive text workloads. Compare it with Sonnet using the same examples to learn which tasks need a different model.

text

Try in Playground ↗

Input / 1M tokens$1

Output / 1M tokens$5

Context window200,000

Maximum output64,000

Provider-supported parameters

These are upstream capabilities. Jevrouter supports the subset described in the API contract; chat input is currently text only.

include_reasoningmax_completion_tokensmax_tokensreasoningresponse_formatstopstructured_outputstemperaturetool_choicetoolstop_ktop_pstreamdeveloper_role

Capabilities are imported from OpenRouter’s catalog. This does not mean every model has been independently tested by Jevrouter.

Catalog updated: 2026-09-22T08:51:50.371Z · Price version: cmubxr28i07t4jn0zc6zdn8iw:3

Provider pricing fields · USD per indicated unit

{
  "prompt": "0.000001",
  "completion": "0.000005",
  "web_search": "0.01",
  "input_cache_read": "0.0000001",
  "input_cache_write": "0.00000125",
  "input_cache_write_1h": "0.000002"
}

JEVROUTER INTEGRATION NOTES · REVIEWED 2026-09-22

What to check in your application

Measure time to first token separately from total completion time in interactive applications.
Inspect the provider receipt rather than assuming cached-input discounts apply to every call.

Start with the example below, then test representative inputs against your own acceptance criteria.

Make a request

curl --no-buffer https://api.jevrouter.io/v1/chat/completions \
  -H "Authorization: Bearer $JEVROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d @- <<'JSON'
{
  "model": "anthropic/claude-haiku-4.5",
  "messages": [
    {
      "role": "user",
      "content": "Rewrite this support response to be concise and polite: You entered the wrong billing address. Change it in settings and try again."
    }
  ],
  "max_tokens": 1024
}
JSON

Plan your integration

Connect with the OpenAI Python SDK →Estimate this model’s cost →Budgets, request status and safe retries →

Related models to evaluate

Use identical inputs and scoring rules. No ranking is implied by this selection.

OpenAI: GPT-4.1 Mini Anthropic: Claude Sonnet 4

Source: OpenRouter model catalog. Imported descriptions and capabilities are provider declarations, not independent benchmark results. Jevrouter currently uses OpenRouter as its managed upstream.