Anthropic Claude Haiku 4.5 — what it is, when to use it and what to use instead

Summary

Claude Haiku 4.5 is the cheapest model in the current Claude line-up, at $1 per million input tokens and $5 per million output tokens, and Anthropic rates it as the fastest of its current models. It is bounded by a 200K-token context window and a February 2025 knowledge cutoff.

Haiku is the small tier of the Claude line. Version 4.5 was released on 15 October 2025 and is still the current Haiku model, which makes it the oldest of the four current Claude models. Its limits reflect that: a 200K-token context window and 64K max output, against 1M and 128K on Sonnet 5, Opus 5 and Fable 5.1.

It accepts text and image input and returns text. Thinking is the manual extended mode rather than the adaptive mode used by the newer models, and the effort parameter is not supported. The reliable knowledge cutoff is February 2025, the oldest of the current Claude models.

No deprecation has been announced. Anthropic commits to keeping Claude Haiku 4.5 available until at least 15 October 2026, which is the nearest commitment date of the four current Claude models.

Modality

• Text
• Vision

Use case

• Chat
• Coding
• Reasoning
• Agents & tool use
• Long context

Tier

• Small

Access

• Hosted API

Languages

• Multilingual

Specifications

Input price

$1.00 / M tokens

Output price

$5.00 / M tokens

Price band

Standard

Context window

200K tokens

Max output

64K tokens

Input

Text, image

Output

Text

Knowledge cutoff

February 2025

Tool use

Yes

Structured output

Yes

Weights licence

not published

Model maker

Anthropic

“unknown” means the figure has not been published, or we have not verified it yet — never that the answer is no. “not published” under the weights licence means this model is not offered with downloadable weights. The price band is ours: it bands the blended price (three parts input to one part output) as budget (≤ $0.50), standard (≤ $3), premium (≤ $12) or flagship.

Lifecycle

Anthropic has not announced a deprecation or shutdown date for this model. That is a statement about what has been published, not a guarantee — providers typically announce a shutdown months ahead, and this page updates when they do.

1. Released 15 October 2025
2. Deprecated not announced
3. Retired not announced

Use it for

• Classify, route or summarise high volumes of short inputs where per-call cost decides the architecture.
• Run sub-agents and tool-calling steps that need the fastest possible turnaround inside a larger workflow.
• Keep a cheap fallback tier alongside Sonnet or Opus for the simple half of your traffic.

Avoid it when

• Your prompts depend on facts after February 2025, which is this model's reliable knowledge cutoff.
• Your inputs exceed 200K tokens or your responses need more than 64K tokens.
• The task needs deep multi-step reasoning; Claude Sonnet 5 is the next tier up at $2 / $10.
• You need audio or video input; Claude Haiku 4.5 accepts text and images only.

Alternatives to consider

• Claude Sonnet 5 Curated

  Anthropic ● Active

  Twice the price at $2 / $10, but a 1M-token window and a January 2026 knowledge cutoff instead of February 2025.

  $2.00 in · $10.00 out / M tokens · 1M context

• GPT-5.6 Luna Curated

  OpenAI ● Active

  OpenAI's cost tier at $0.2 / $1.2, a fifth of the input price, with a 1.05M-token context window.

  $0.20 in · $1.20 out / M tokens · 1.1M context

• Gemini 2.5 Flash-Lite Curated

  Google (Gemini) ● Active

  Google's cheapest listed model at $0.1 / $0.4, with roughly a 1M-token window and audio, video and PDF input.

  $0.10 in · $0.40 out / M tokens · 1M context

• Gemini 3.5 Flash-Lite Curated

  Google (Gemini) ● Active

  A newer low-cost option at $0.3 / $2.5 with roughly a 1M-token window, if the 200K limit is what blocks you.

  $0.30 in · $2.50 out / M tokens · 1M context

• Claude Haiku 4.5 Derived

  Amazon Bedrock ● Active

  Alternative from Amazon Bedrock, same tier (small), similar price; shares chat, coding, reasoning, agents and long-context.

  $1.00 in · $5.00 out / M tokens · 200K context

Curated suggestions were picked by a reviewer, with the reason written by hand. Derived suggestions are ranked from the data — shared tags, the same tier and a comparable price band — not from benchmark scores. Check the capability fit yourself: we track lifecycle, price and specifications.

Sources

• platform.claude.com ↗ https://platform.claude.com/docs/en/models/haiku-4-5/overview
• platform.claude.com ↗ https://platform.claude.com/docs/en/about-claude/pricing
• platform.claude.com ↗ https://platform.claude.com/docs/en/about-claude/models/overview
• platform.claude.com ↗ https://platform.claude.com/docs/en/about-claude/model-deprecations

Verified against the source on 10 September 2026. Dataset snapshot 10 September 2026. Providers can change a date without notice — the link above is always the final word.

About this profile

The specifications and dates above are facts from Anthropic, with the sources listed. The summary, the use-for and avoid-when lists and the curated alternatives are our assessment: editorial profile reviewed on 21 September 2026 by llm-researcher.