Claude Haiku 4.5

Haiku 4.5 is the Claude you put in front of users when they are waiting for an answer, or behind a bigger model when you need many small jobs done at once. Anthropic released it in October 2025 as its small model, and at launch compared it with Claude Sonnet 4: similar coding level for roughly a third of the price and over twice the speed, plus better scores on some computer-use tasks.

On Kadraris it runs from the same key and balance as the larger Claude models, which makes mixed setups easy to price.

What you can use it for

• Support chat and in-app assistants where a slow reply loses the user
• Worker agents that take small subtasks from a planner model in parallel
• Sorting, tagging and pulling fields out of thousands of records
• Fast back-and-forth prototyping inside Claude Code

Settings and limits

• Context window: 200,000 tokens. That holds a long chat history plus a decent slice of a help center.
• Output: up to 64K tokens per reply.
• For whole repositories or large document piles, move up to a 1M-token model such as Claude Sonnet 5.
• Keep prompts short and give it a fixed answer format; small models follow tight instructions best.
• Same endpoints as other Claude models: /v1/chat/completions and /v1/messages.

Example request

from openai import OpenAI

client = OpenAI(
    base_url="https://kadraris.com/v1",
    api_key="sk-kd-...",
)
resp = client.chat.completions.create(
    model="claude-haiku-4.5",
    messages=[{"role": "user",
               "content": "Label this email as refund, "
                          "shipping or other. One word."}],
)
print(resp.choices[0].message.content)

Replace the labels with your own categories and keep the one-word rule for clean parsing.

Questions

What will a high-volume Haiku job cost?

Haiku 4.5 is billed at $1.35 / $6.75 per 1M tokens, the lowest rate among the Claude models here. Multiply that by your average tokens per call to estimate a daily run. Calls that fail are not charged.

Can I run a planner plus several Haiku workers?

Yes, and Anthropic recommends that pattern: a larger Claude splits the job and Haiku 4.5 copies handle the pieces. With Kadraris every call uses one key, so the full cost of the setup shows up in a single usage view.

Does it work in Claude Code?

It does. Point Claude Code at Kadraris as described in the Claude Code guide and select claude-haiku-4.5.

Claude Haiku 4.5 in your code

from openai import OpenAI

client = OpenAI(base_url="https://kadraris.com/v1", api_key="sk-kd-...")
resp = client.chat.completions.create(
    model="claude-haiku-4.5",
    messages=[{"role": "user", "content": "Suggest three opening shots for a short film about a cafe."}],
)
print(resp.choices[0].message.content)

curl https://kadraris.com/v1/chat/completions \
  -H "Authorization: Bearer $KADRARIS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "claude-haiku-4.5", "messages": [{"role": "user", "content": "Suggest three opening shots for a short film about a cafe."}]}'

import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://kadraris.com/v1", apiKey: process.env.KADRARIS_API_KEY });
const resp = await client.chat.completions.create({
  model: "claude-haiku-4.5",
  messages: [{ role: "user", content: "Suggest three opening shots for a short film about a cafe." }],
});
console.log(resp.choices[0].message.content);

export ANTHROPIC_BASE_URL=https://kadraris.com
export ANTHROPIC_AUTH_TOKEN=sk-kd-...
export ANTHROPIC_MODEL=claude-haiku-4.5
claude