Claude Haiku 4.5

Claude Haiku 4.5 is the volume model. At $1 in and $5 out per million tokens it is a fifth the price of Opus 5 on input, and Anthropic’s framing at launch was that it delivers coding performance similar to Claude Sonnet 4 at a third the cost and more than twice the speed.

It is also the odd one out in the current line-up in two ways.

Two ways it differs from the Claude 5 models

It still has the old context window. 200K tokens, not 1M. If your workload depends on very long context, Haiku 4.5 is not a drop-in substitute for Sonnet 5 — and given Sonnet 5 is only twice the price, the calculation often favours Sonnet.

It uses the old thinking model. Haiku 4.5 supports thinking.type: "enabled" — explicit extended thinking — and does not support adaptive thinking. Every Claude 5 model is the reverse. Code that targets both needs to branch.

It also still uses the pre-4.7 tokenizer, so the same text produces roughly 30% fewer tokens on Haiku 4.5 than on Fable 5 or Opus 4.7 and later. That difference cuts the other way from what the per-token price suggests: comparing raw rates across the tokenizer boundary understates Haiku’s cost advantage.

Specifications

Property	Value
Claude API ID	claude-haiku-4-5-20251001
Claude API alias	claude-haiku-4-5
Amazon Bedrock ID	anthropic.claude-haiku-4-5-20251001-v1:0
Google Cloud ID	claude-haiku-4-5@20251001
Context window	200K tokens
Max output	64K tokens
Extended thinking parameter	Yes
Adaptive thinking	No
Reliable knowledge cutoff	February 2025
Training data cutoff	July 2025

Pricing

Rate	Price per million tokens
Input	$1
Output	$5
5-minute cache write	$1.25
1-hour cache write	$2
Cache read	$0.10
Batch input / output	$0.50 / $2.50

Combining batch processing with prompt caching puts effective input cost at $0.05 per million tokens on cache hits.

Lifecycle

Active, tentative retirement not sooner than 15 October 2026 — the nearest retirement date of any current model, and worth watching if you have Haiku 4.5 in production.