Claude Haiku 5.5 API: the fast, low-cost Claude with a 1M context

Claude Haiku 5.5 is the smallest and fastest model of Anthropic's new Claude generation. On GetVibe API it is available as claude-haiku-5.5 through the OpenAI-compatible /v1/chat/completions endpoint, the Anthropic Messages endpoint /v1/messages, and agent tools such as Claude Code. You pay per token from your USD balance, with the same key you use for every other model.

Specs

What it is good at

Haiku is for work where requests are many, each one is simple, and the answer has to come back fast. It keeps the Claude habits people choose it for: it follows instructions closely and returns the format you asked for.

• Classification and routing — support tickets, review sentiment, moderation, lead scoring.
• Extraction — fields from emails, contracts, invoices and forms into JSON, including from a scan or a screenshot.
• Support bots and FAQ answers over a knowledge base, where speed and a steady tone matter more than deep reasoning.
• Agent sub-steps — summarizing a tool result, checking a format, choosing the next step. Keep the expensive model for planning and hand the routine steps to Haiku.
• Batch jobs — rewriting product descriptions, short summaries, translation of short texts.

What we checked before writing this

All of the following ran live through getvibeapi.com on October 22, 2026:

• Classification. A ticket "I was charged twice for my October top-up" with a four-label system prompt came back as billing. The request used 55 input and 4 output tokens.
• Image input. Given a picture of an invoice number, the model read INV-2291 exactly.
• Tool calling. Asked about an order and a shipping quote at once, it called both functions in one turn and then answered from their results. The full loop is in our function calling guide.
• Prompt caching. We sent the same system prompt twice in a row and read the usage fields of each response.

Very short prompts are not cached. Above about a thousand tokens, the repeat request reads almost the whole prefix from the cache. To get that, keep the part that does not change (instructions, reference text, examples) at the start of the prompt and put the variable data at the end. For Claude models GetVibe places the cache_control breakpoints for you, on /v1/chat/completions too. See prompt caching in the docs.

Haiku 5.5, Sonnet 5.5 or Opus 5.5?

All three have a 1M-token context, image input and reasoning, so the choice comes down to how hard the task is and what it costs:

• Claude Haiku 5.5 — $0.0267 input · $0.1334 output per 1M tokens. High-volume, short tasks: classification, extraction, knowledge-base answers, agent sub-steps.
• Claude Sonnet 5.5 — $0.0534 input · $0.2667 output per 1M tokens. Everyday development, agentic coding, terminal work, long chains of actions.
• Claude Opus 5.5 — $0.1067 input · $0.5334 output per 1M tokens. The hardest problems with long reasoning, where a wrong answer costs more than the tokens.

A good habit is to start with Haiku and move to a larger model only where the answers fall short. The calls are identical; only the model string changes.

Call it

OpenAI-compatible format:

curl https://getvibeapi.com/v1/chat/completions \
  -H "Authorization: Bearer $GETVIBE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-haiku-5.5",
    "max_tokens": 20,
    "messages": [
      {"role": "system", "content": "Classify the support ticket as one of: billing, bug, feature_request, other. Reply with the label only."},
      {"role": "user", "content": "I was charged twice for my October top-up."}
    ]
  }'

Python with the OpenAI SDK:

import os
from openai import OpenAI

client = OpenAI(base_url="https://getvibeapi.com/v1", api_key=os.environ["GETVIBE_API_KEY"])

resp = client.chat.completions.create(
    model="claude-haiku-5.5",
    messages=[{"role": "user", "content": "Extract the vendor, date and total from this invoice text: ..."}],
)
print(resp.choices[0].message.content)

Anthropic Messages format, for the Anthropic SDK and Claude Code:

curl https://getvibeapi.com/v1/messages \
  -H "x-api-key: $GETVIBE_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-haiku-5.5",
    "max_tokens": 512,
    "messages": [{"role": "user", "content": "Summarize this ticket in one line: ..."}]
  }'

To run Claude Code on Haiku for cheap sub-tasks, see the Claude Code guide. Live prices and channels are on the model page, and the price calculator estimates your monthly bill.