What is the Claude Haiku 5.5 100K-token price cliff?

Haiku 5.5 picks its price tier per request, based on the total prompt size. Cross 100,000 prompt tokens and every token in that request — input, cache and output included — is billed at the 5x long-context rates. It is not a marginal bracket like income tax; one token over the line re-prices the entire request.

How does the Batch API discount work for Haiku 5.5?

The Claude Batch API processes requests asynchronously, typically within 24 hours, at 50% off input and output tokens. On Haiku 5.5 that means $0.05 / $0.25 per million tokens for prompts up to 100K, and $0.25 / $1.25 above 100K. Cache pricing applies as usual.

Why does the same text use more tokens on Haiku 5.5 than on Haiku 4.5?

Haiku 5.5 ships with a new tokenizer. Anthropic's migration guide says the same input text produces approximately 30% more tokens on Haiku 5.5 than on Haiku 4.5, so token counts, max_tokens limits and cost estimates measured on 4.5 need to be redone.

How much cheaper is Claude Haiku 5.5 than Haiku 4.5?

On short prompts Haiku 5.5 is 90% cheaper than Haiku 4.5 ($0.10 vs $1.00 input per million). Above 100K prompt tokens it is 50% cheaper ($0.50 vs $1.00). Anthropic puts the blended saving on real workloads at about 75% once request mix and the new tokenizer are accounted for.

Claude Haiku 5.5 Pricing Calculator – API Cost per Request & Month

Q: How much does Claude Haiku 5.5 cost?

Claude Haiku 5.5 has two price tiers based on prompt length. For prompts up to 100,000 tokens it costs $0.10 per million input tokens and $0.50 per million output tokens. For prompts over 100,000 tokens the whole request bills at five times that: $0.50 input and $2.50 output per million. Cache reads cost $0.01 or $0.05 per million depending on tier.

Q: Do cached tokens count toward the 100K threshold?

Anthropic has not publicly clarified whether cache reads and cache writes count toward the 100,000-token threshold. This calculator budgets conservatively and counts all prompt tokens (fresh input + cache writes + cache reads) toward the threshold, so your estimate will not come in low.

Calculate Your Cost

Workload presets

Fresh input tokens per request

Output tokens per request

Cache write tokens per request

Cache read tokens per request

Requests per month

Cache write TTL

Batch API (50% off input & output, async)

Tier rule used here: total prompt tokens = fresh input + cache writes + cache reads. Anthropic has not clarified whether cached tokens count toward the 100K line, so this calculator budgets as if they do.

Your Estimate

Standard tier

Input$0.000000

Cache write$0.000000

Cache read$0.000000

Output$0.000000

Cost per request: $0.00

Estimated monthly cost: $0.00

Claude Haiku 5.5 Rate Card (per 1M tokens)

Token type	Haiku 5.5 ≤100K	Haiku 5.5 >100K	Haiku 4.5	Sonnet 5.5
Input	$0.10	$0.50	$1.00	$2.00
Output	$0.50	$2.50	$5.00	$10.00
Cache write, 5 min	$0.125	$0.625	$1.25	$2.50
Cache write, 1 hour	$0.20	$1.00	$2.00	$4.00
Cache read	$0.01	$0.05	$0.10	$0.10
Batch input	$0.05	$0.25	$0.50	$1.00
Batch output	$0.25	$1.25	$2.50	$5.00

Rates from Anthropic's official Haiku 5.5 launch post and pricing page, verified October 20, 2026. Haiku 5.5: 1M-token context window, up to 128K output tokens per request, model ID claude-haiku-5-5.

The 100K Cliff: What Actually Drives Your Bill

The tier is chosen per request, and it re-prices everything. A 100,001-token prompt does not just pay more for the extra token — input, cache and output for that whole request bill at 5× the base rates.
Output price follows the prompt's tier. The length of what you send decides the price of what comes back: the same 1,000-token answer costs $0.0005 under the line and $0.0025 over it.
Budget cached tokens as counting toward the line. Anthropic has not clarified this either way. A giant cached system prompt may save on reads and still push the request into the 5× tier — keep total prompt size in view, not just fresh input.
Recount your tokens. Haiku 5.5's new tokenizer produces roughly 30% more tokens for the same text than Haiku 4.5's. Every estimate, limit and budget carried over from 4.5 is stale.
Stay under the line by compacting, not by hoping. Trim conversation history, summarize tool output, and move static context into cache — then check the total prompt size in this calculator before you ship.
Stack the boring discounts. Batch API halves input and output for anything that can wait up to 24 hours, and cache reads cost a tenth of fresh input on short prompts. Combined, routine workloads run at a small fraction of list price.

FAQ

How much does Claude Haiku 5.5 cost?

Two tiers, set by prompt length: up to 100,000 prompt tokens, $0.10 per million input and $0.50 per million output; over 100,000, the whole request bills at $0.50 / $2.50. Cache reads are $0.01 or $0.05 per million by tier.

What is the 100K-token price cliff?

Haiku 5.5 picks a price tier per request from the total prompt size. Cross 100K tokens and every token in that request — input, cache and output — bills at the 5× long-context rates. It is a cliff, not a marginal bracket.

Do cached tokens count toward the 100K threshold?

Anthropic has not publicly clarified this. This calculator counts all prompt tokens (fresh input + cache writes + cache reads) toward the threshold, so the estimate errs on the safe side.

How does the Batch API discount work?

Batch requests run asynchronously, typically within 24 hours, at 50% off input and output tokens: $0.05 / $0.25 per million under 100K, $0.25 / $1.25 over it. Cache pricing applies as usual.

Why does the same text use more tokens than on Haiku 4.5?

Haiku 5.5 uses a new tokenizer that produces approximately 30% more tokens for the same text. Recount prompts before rebudgeting, and re-check your max_tokens limits.

How much cheaper is Haiku 5.5 than Haiku 4.5?

90% cheaper on short prompts, 50% cheaper on long ones. Anthropic's blended estimate for real workloads is about 75% average savings after request mix and the tokenizer change.

Methodology & sources: rates verified against Anthropic's official launch post and pricing page on October 20, 2026 (anthropic.com/claude-haiku-5-5). Estimates exclude server-side tool fees (e.g. web search at $10 per 1,000 calls), the 1.1× US-only inference multiplier, and taxes. This is an estimate, not a quote — your invoice follows Anthropic's metered usage.