Claude Haiku 4.5: specs, price and best uses

Claude Haiku 4.5 specs

The table below covers the core limits, modalities and model identifiers for claude-haiku-4-5-20251001 and its alias claude-haiku-4-5.

• The 200K-token context window holds roughly 150k words.
• Extended thinking is manual (thinking.type: "enabled" with budget_tokens), and the model does not accept the effort parameter.
• Training data cuts off at July 2025, with reliable knowledge through February 2025.

What Haiku 4.5 is good at

• Real-time, low-latency apps like chat assistants, customer service agents and pair programming.
• Sub-agent setups where a larger model like Sonnet 4.5 plans the work and several Haiku 4.5 instances run the subtasks in parallel.
• Computer use tasks, where Anthropic said it beat Sonnet 4.
• Quick work in Claude Code, such as rapid prototyping and multi-agent projects.

In Anthropic's October 15, 2025 launch post, Haiku 4.5 scored 73.3% on SWE-bench Verified (averaged over 50 trials on the 500-problem set with a 128K thinking budget) and 41.75% on Terminal-Bench with a 32K thinking budget, or 40.21% without thinking. Anthropic describes its coding performance as similar to Claude Sonnet 4 at one-third the cost and more than twice the speed.

Haiku 4.5 pricing with a cost example

The table below lists Claude Haiku 4.5's standard, prompt caching and Batch API rates, and you can compare them with the rest of the lineup on our Claude API pricing page.

Say you run 1,000 requests that each send 2,000 input tokens and return 500 output tokens, which comes out to 2M input tokens and 0.5M output tokens in total. At $1 per million input tokens and $5 per million output tokens, that batch costs $2 + $2.50 = $4.50 on standard billing. If the workload can wait for the Batch API's 50% discount, the same run drops to $2.25.

Haiku 4.5 vs Haiku 5.5

If you are deciding between Haiku 4.5 and Haiku 5.5 (claude-haiku-5-5, released October 7, 2026), the table below lines up their limits, cutoffs and token prices side by side.

• Haiku 5.5's rate goes from $0.10 / $0.50 per million tokens up to $0.50 input and $2.50 output once a prompt passes 100,000 tokens.
• Haiku 5.5 uses Claude's newer tokenizer, which counts the same text as about 30% more tokens than Haiku 4.5.
• Even with 30% more tokens (2.6M input and 0.65M output), the 1,000-request example from the pricing section costs about $0.26 + $0.33 = $0.59 on Haiku 5.5, compared with $4.50 on Haiku 4.5.

Availability and retirement

The table below lists the exact model string to pass on the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry.

Haiku 4.5's status is Active (legacy), which means it still works but gets no further updates. Anthropic's docs list its retirement floor as "not sooner than October 15, 2026," and as of October 29, 2026 that date has passed with no deprecation date announced; when Anthropic does schedule a retirement for a publicly released model, it gives at least 60 days notice before requests start failing.

In the consumer Claude apps, Claude's pricing page lists Haiku models as included on the Free, Pro and Max plans. If your workload needs a larger 1M-context model instead, see our guide to Claude Sonnet 5.

Run Haiku 4.5 on your Claude subscription

Plan2API connects a Claude Pro or Max subscription and exposes claude/claude-haiku-4-5 (200K context, 64K max output) through an OpenAI-compatible Chat Completions endpoint at https://api.plan2api.com/v1 (POST /v1/chat/completions), using an Authorization: Bearer header with your p2a_ key. Instead of paying a per-token API bill, your high-volume Haiku calls run against the usage limits of the Claude subscription you already pay for, and you can also pass the anthropic/ prefix if your code already uses that convention.

The Plan2API Free plan costs $0 with no card required and gives you 1 connected subscription, 2 API keys and 50 chat requests a day, while Pro is $12 a month (or $99 a year) with 3 connected subscriptions and no daily request cap; see the Claude subscription API guide or connect your account at plan2api.com/connect. A Node example calling claude/claude-haiku-4-5 follows below.

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.plan2api.com/v1",
  apiKey: process.env.PLAN2API_KEY,
});

const completion = await client.chat.completions.create({
  model: "claude/claude-haiku-4-5",
  messages: [{ role: "user", content: "Classify this ticket as billing, bug or feature request." }],
});

console.log(completion.choices[0].message.content);