At or below 100K
- Input: $0.10 / MTok
- Output: $0.50 / MTok
- Cache read: $0.01 / MTok
- Best unit economics for short, high-volume work
API PRICING GUIDE · VERIFIED OCTOBER 11, 2026
The complete cost guide for both prompt-length tiers, cache reads and writes, Amazon Bedrock, Batch API and regional inference—with worked examples you can audit.
For prompts over 100,000 tokens, Haiku 5.5 rises to $0.50 per million input tokens and $2.50 per million output tokens. The threshold is evaluated per request, not from your monthly token total.
USD per one million tokens on the Claude Platform.
| Token category | Prompt ≤100K | Prompt >100K |
|---|---|---|
| Input | $0.10 | $0.50 |
| Output | $0.50 | $2.50 |
| Cache read / hit | $0.01 | $0.05 |
| Cache write · 5 minutes | $0.125 | $0.625 |
| Cache write · 1 hour | $0.20 | $1.00 |
The short-prompt tier is one fifth of the long-prompt tier for each token category. Cache hits are one tenth of the tier’s base input rate.
Open the interactive calculatorIt is not a monthly volume threshold.
Anthropic describes the lower tier as prompts “up to 100,000 tokens” and the higher tier as prompts “over 100,000 tokens.” Exactly 100,000 prompt tokens therefore remain in the lower tier.
Standard global pricing, before any cache or Batch adjustments.
10K input + 1K output in the short tier: 10,000 × $0.10/MTok + 1,000 × $0.50/MTok.
100K input + 20K output in the short tier. At 1,000 requests, that is $20.
200K input + 50K output in the long tier. At 1,000 requests, that is $225.
request cost = (input ÷ 1M × input rate) + (output ÷ 1M × output rate) + cache costsMultiply the request result by request volume. Add any paid server-side tool usage separately.
Pricing modifiers may stack, so model the full request configuration.
Batch API discount on input and output tokens for asynchronous processing.
Haiku cache-hit multiplier versus the applicable base input rate.
US-only inference multiplier for Claude 4.6 and later models; global routing uses standard pricing.
A five-minute cache write costs 1.25× the base input rate and a one-hour write costs 2×. Caching pays off when reused prompt content avoids repeated full-price input processing.
Amazon Bedrock bills Haiku 5.5 through AWS Marketplace. These figures are USD per one million tokens and were checked against AWS pricing data on October 11, 2026.
| Token category | Global ≤100K | US Geo ≤100K | Global >100K | US Geo >100K |
|---|---|---|---|---|
| Input | $0.10 | $0.11 | $0.50 | $0.55 |
| Output | $0.50 | $0.55 | $2.50 | $2.75 |
| Cache read | $0.01 | $0.011 | $0.05 | $0.055 |
| Cache write · 5 min | $0.125 | $0.1375 | $0.625 | $0.6875 |
| Cache write · 1 hour | $0.20 | $0.22 | $1.00 | $1.10 |
anthropic.claude-haiku-5-5global.anthropic.claude-haiku-5-5us.anthropic.claude-haiku-5-5Amazon Bedrock uses Bedrock model or inference-profile IDs and AWS-managed features. Claude Platform on AWS provides Anthropic’s native API and console experience with AWS billing and authentication; its model ID remains claude-haiku-5-5.
Prompts up to 100K cost $0.10 per million input tokens and $0.50 per million output tokens. Prompts over 100K cost $0.50 input and $2.50 output at the Global base rate.
Yes. Anthropic’s rate card says “up to 100,000 tokens” for the lower tier and “over 100,000 tokens” for the higher tier.
In the Global short tier, a five-minute write is $0.125/MTok, a one-hour write is $0.20/MTok and a cache hit is $0.01/MTok. In the long tier they are $0.625, $1.00 and $0.05 respectively.
On the Claude Platform, yes: Anthropic documents a 50% input and output discount. The current Amazon Bedrock model card marks Batch unsupported for Haiku 5.5, so do not apply the discount to Bedrock workloads.
Global cross-Region inference uses the base rates. US Geo cross-Region inference is 1.1×: for prompts up to 100K, that is $0.11 input and $0.55 output per million tokens; above 100K, $0.55 input and $2.75 output.
No. Bedrock is AWS-managed model access with Bedrock IDs and features. Claude Platform on AWS is Anthropic’s native API and console experience with AWS billing and authentication.
Use claude-haiku-5-5 on the Claude Platform and anthropic.claude-haiku-5-5 on Bedrock. Cross-Region Bedrock profiles add a scope prefix such as global. or us..