Claude Haiku 5.5 is now available on Amazon Bedrock
This page has been translated by machine translation. View original
Introduction
On October 7, 2026, Claude Haiku 5.5 became available on Amazon Bedrock.
In this article, I called Haiku 5.5 via AWS CLI converse using IAM authentication. I also checked how it appears in CloudTrail event history and confirmed the unit prices retrieved from the Price API. The unit prices are compared with Sonnet 5.5, Opus 5.5, and others.
Haiku 5.5 Context Length and Inference Profiles
According to the model card, the Context window is 200K tokens for Haiku 4.5 and 1M tokens for Haiku 5.5. Sonnet 5.5 and Opus 5.5 are also 1M tokens.
The model ID for Haiku 5.5 is anthropic.claude-haiku-5-5, and inference profiles with the prefixes us, eu, au, jp, and global are available.
Calling converse with IAM Authentication
The sample code in the model card describes a procedure for issuing a long-term API key from the Bedrock console and calling it from Python. In this article, I called it using IAM role session credentials without using an API key.
JP Profile
This is the command specifying the JP inference profile.
aws bedrock-runtime converse --region ap-northeast-1 \
--model-id jp.anthropic.claude-haiku-5-5 \
--messages '[{"role":"user","content":[{"text":"Bedrock の Claude Haiku 5.5 について、一文で自己紹介してください。"}]}]' \
--inference-config '{"maxTokens":1000}'
Here is an excerpt of the response. The reasoningContent block and metrics are omitted.
{
"output": {
"message": {
"role": "assistant",
"content": [
{
"text": "私はAnthropicが開発したAIアシスタントのClaudeで、Amazon Bedrockから利用されているClaude Haiku 5.5として、質問への回答、文章作成、要約、翻訳、プログラミングなどのお手伝いをします。"
}
]
}
},
"stopReason": "end_turn",
"usage": {
"inputTokens": 39,
"outputTokens": 469,
"totalTokens": 508,
"cacheReadInputTokens": 0,
"cacheWriteInputTokens": 0
}
}
The text in the omitted reasoningContent was an empty string. However, outputTokens is a larger value than the character count of the response body. The model card states that adaptive thinking is enabled by default. Output costs are estimated based on usage outputTokens, not the length of the response body.
Global Profile
I ran the same prompt with --model-id changed to global.anthropic.claude-haiku-5-5. The exit code was 0, stopReason was end_turn, and usage was inputTokens 39, outputTokens 335, totalTokens 374. As with JP, the text in reasoningContent was an empty string.
How It Appears in CloudTrail Event History
I checked the event history using lookup-events without creating additional trails or S3 buckets.
aws cloudtrail lookup-events --region ap-northeast-1 \
--start-time 2026-10-08T05:44:00+09:00 --end-time 2026-10-08T05:56:00+09:00 \
--lookup-attributes AttributeKey=EventSource,AttributeValue=bedrock.amazonaws.com
When I checked approximately 10 minutes after the calls, both the JP and Global entries were displayed. The event source is bedrock.amazonaws.com, not bedrock-runtime. A search in a different time range (05:25–05:40) also included ListGuardrails and ListTagsForResource from the AWS Config service-linked role, so I narrowed the results to events where the event name is Converse.
CloudTrailEvent is returned as a JSON string. The following is an excerpt from the expanded event for the Global call. Identifiers such as access keys and role names have been removed.
{
"EventName": "Converse",
"ReadOnly": "true",
"EventTime": "2026-10-08T05:45:27+09:00",
"EventSource": "bedrock.amazonaws.com",
"CloudTrailEvent": {
"eventTime": "2026-10-07T20:45:27Z",
"eventSource": "bedrock.amazonaws.com",
"eventName": "Converse",
"awsRegion": "ap-northeast-1",
"requestParameters": {
"modelId": "global.anthropic.claude-haiku-5-5",
"inferenceConfig": {
"maxTokens": 1000
}
},
"responseElements": null,
"additionalEventData": {
"inferenceRegion": "us-east-1",
"inputTokens": 39,
"outputTokens": 335
},
"readOnly": true,
"resources": [
{
"accountId": "<ACCOUNT_ID>",
"type": "AWS::Bedrock::Model",
"ARN": "arn:aws:bedrock:ap-northeast-1::foundation-model/anthropic.claude-haiku-5-5"
}
],
"eventType": "AwsApiCall",
"managementEvent": true,
"eventCategory": "Management"
}
}
The model called can be identified from modelId in requestParameters. The prompt body (messages) is not included in requestParameters.
The processing region can be identified from inferenceRegion in additionalEventData. In this case, the JP call processed in ap-northeast-1 and the Global call in us-east-1. The ARN in resources pointed to the foundation model in the calling region for both JP and Global calls, so it cannot be used to determine the processing region. The inputTokens and outputTokens in additionalEventData matched the values in the response usage.
Pricing
Checking Unit Prices via the Price API
aws pricing get-products --region us-east-1 --service-code AmazonBedrockFoundationModels \
--filters 'Type=TERM_MATCH,Field=servicename,Value=Claude Haiku 5.5 (Amazon Bedrock Edition)' \
'Type=TERM_MATCH,Field=regionCode,Value=us-east-1'
Execution result (excerpt)
{
"PriceList": [
"{\"product\":{\"attributes\":{\"regionCode\":\"us-east-1\",\"usagetype\":\"USE1-MP:USE1_output_tokens_global_standard-Units\",\"locationType\":\"AWS Region\",\"location\":\"US East (N. Virginia)\",\"servicename\":\"Claude Haiku 5.5 (Amazon Bedrock Edition)\",\"operation\":\"\"},\"sku\":\"9C6E7ATQE23Q4W66\"},\"serviceCode\":\"AmazonBedrockFoundationModels\",\"terms\":{\"OnDemand\":{\"9C6E7ATQE23Q4W66.4799GE89SK\":{\"priceDimensions\":{\"9C6E7ATQE23Q4W66.4799GE89SK.6YS6EN2CT7\":{\"unit\":\"1M tokens\",\"endRange\":\"Inf\",\"description\":\"AWS Marketplace software usage|us-east-1|Output Tokens - Standard, Global\",\"appliesTo\":[],\"rateCode\":\"9C6E7ATQE23Q4W66.4799GE89SK.6YS6EN2CT7\",\"beginRange\":\"0\",\"pricePerUnit\":{\"USD\":\"0.5000000000\"}}},\"sku\":\"9C6E7ATQE23Q4W66\",\"effectiveDate\":\"2026-10-01T00:00:00Z\",\"offerTermCode\":\"4799GE89SK\",\"termAttributes\":{}}}},\"version\":\"20261007183733\",\"publicationDate\":\"2026-10-07T18:37:33Z\"}",
"{\"product\":{\"attributes\":{\"regionCode\":\"us-east-1\",\"usagetype\":\"USE1-MP:USE1_input_tokens_global_standard-Units\",\"locationType\":\"AWS Region\",\"location\":\"US East (N. Virginia)\",\"servicename\":\"Claude Haiku 5.5 (Amazon Bedrock Edition)\",\"operation\":\"\"},\"sku\":\"CT2JY6DA5N94ZAXN\"},\"serviceCode\":\"AmazonBedrockFoundationModels\",\"terms\":{\"OnDemand\":{\"CT2JY6DA5N94ZAXN.4799GE89SK\":{\"priceDimensions\":{\"CT2JY6DA5N94ZAXN.4799GE89SK.6YS6EN2CT7\":{\"unit\":\"1M tokens\",\"endRange\":\"Inf\",\"description\":\"AWS Marketplace software usage|us-east-1|Input Tokens - Standard, Global\",\"appliesTo\":[],\"rateCode\":\"CT2JY6DA5N94ZAXN.4799GE89SK.6YS6EN2CT7\",\"beginRange\":\"0\",\"pricePerUnit\":{\"USD\":\"0.1000000000\"}}},\"sku\":\"CT2JY6DA5N94ZAXN\",\"effectiveDate\":\"2026-10-01T00:00:00Z\",\"offerTermCode\":\"4799GE89SK\",\"termAttributes\":{}}}},\"version\":\"20261007183733\",\"publicationDate\":\"2026-10-07T18:37:33Z\"}",
"{\"product\":{\"attributes\":{\"regionCode\":\"us-east-1\",\"usagetype\":\"USE1-MP:USE1_input_tokens_long_ctx_global_standard-Units\",\"locationType\":\"AWS Region\",\"location\":\"US East (N. Virginia)\",\"servicename\":\"Claude Haiku 5.5 (Amazon Bedrock Edition)\",\"operation\":\"\"},\"sku\":\"E7TVM9F3MW6M99X9\"},\"serviceCode\":\"AmazonBedrockFoundationModels\",\"terms\":{\"OnDemand\":{\"E7TVM9F3MW6M99X9.4799GE89SK\":{\"priceDimensions\":{\"E7TVM9F3MW6M99X9.4799GE89SK.6YS6EN2CT7\":{\"unit\":\"1M tokens\",\"endRange\":\"Inf\",\"description\":\"AWS Marketplace software usage|us-east-1|Input Tokens - Standard, Long Context, Global\",\"appliesTo\":[],\"rateCode\":\"E7TVM9F3MW6M99X9.4799GE89SK.6YS6EN2CT7\",\"beginRange\":\"0\",\"pricePerUnit\":{\"USD\":\"0.5000000000\"}}},\"sku\":\"E7TVM9F3MW6M99X9\",\"effectiveDate\":\"2026-10-01T00:00:00Z\",\"offerTermCode\":\"4799GE89SK\",\"termAttributes\":{}}}},\"version\":\"20261007183733\",\"publicationDate\":\"2026-10-07T18:37:33Z\"}",
"{\"product\":{\"attributes\":{\"regionCode\":\"us-east-1\",\"usagetype\":\"USE1-MP:USE1_cache_read_tokens_long_ctx_standard-Units\",\"locationType\":\"AWS Region\",\"location\":\"US East (N. Virginia)\",\"servicename\":\"Claude Haiku 5.5 (Amazon Bedrock Edition)\",\"operation\":\"\"},\"sku\":\"XCA8KP4W9DAXTNDB\"},\"serviceCode\":\"AmazonBedrockFoundationModels\",\"terms\":{\"OnDemand\":{\"XCA8KP4W9DAXTNDB.4799GE89SK\":{\"priceDimensions\":{\"XCA8KP4W9DAXTNDB.4799GE89SK.6YS6EN2CT7\":{\"unit\":\"1M tokens\",\"endRange\":\"Inf\",\"description\":\"AWS Marketplace software usage|us-east-1|Cache Read Tokens - Standard, Long Context\",\"appliesTo\":[],\"rateCode\":\"XCA8KP4W9DAXTNDB.4799GE89SK.6YS6EN2CT7\",\"beginRange\":\"0\",\"pricePerUnit\":{\"USD\":\"0.0550000000\"}}},\"sku\":\"XCA8KP4W9DAXTNDB\",\"effectiveDate\":\"2026-10-01T00:00:00Z\",\"offerTermCode\":\"4799GE89SK\",\"termAttributes\":{}}}},\"version\":\"20261007183733\",\"publicationDate\":\"2026-10-07T18:37:33Z\"}"
],
"FormatVersion": "aws_v1"
}
Haiku 5.5 Unit Prices (Global / US / JP)
The unit is USD / 1M tokens, based on the Price List published at 2026-10-07T18:37:33Z. The US column is for us-east-1 Standard, and the JP column is for ap-northeast-1 Standard. The Global column had the same values for both us-east-1 and ap-northeast-1. The "Prompt over 100K" rows show the unit prices when the prompt exceeds 100K tokens.
| Item | Global | US | JP |
|---|---|---|---|
| Input | 0.10 | 0.11 | 0.11 |
| Output | 0.50 | 0.55 | 0.55 |
| Cache write 5 min | 0.125 | 0.1375 | 0.1375 |
| Cache write 1 hour | 0.20 | 0.22 | 0.22 |
| Cache read | 0.01 | 0.011 | 0.011 |
| Input (prompt over 100K) | 0.50 | 0.55 | 0.55 |
| Output (prompt over 100K) | 2.50 | 2.75 | 2.75 |
| Cache write 5 min (prompt over 100K) | 0.625 | 0.6875 | 0.6875 |
| Cache write 1 hour (prompt over 100K) | 1.00 | 1.10 | 1.10 |
| Cache read (prompt over 100K) | 0.05 | 0.055 | 0.055 |
Comparing the us-east-1 and ap-northeast-1 values, US and JP are identical across all items, and both are 1.1 times the Global price.
Comparison with Haiku 4.5, Sonnet 5.5, and Opus 5.5
According to Anthropic's pricing page, models from Claude 4.6 onward (except Haiku 5.5) can use 1M tokens of context at the standard price. For Haiku 5.5, pricing is determined by prompt length, and exceeding 100K tokens results in a higher price. Sonnet 5.5 and Opus 5.5 have a flat rate up to 1M tokens.
The following is a comparison of key items for Global (USD / 1M tokens, us-east-1).
| Model | Context | Input | Output | Cache write 5 min | Cache read |
|---|---|---|---|---|---|
| Haiku 4.5 | 200K | 1.00 | 5.00 | 1.25 | 0.10 |
| Haiku 5.5 (prompt 100K or less) | 1M | 0.10 | 0.50 | 0.125 | 0.01 |
| Haiku 5.5 (prompt over 100K) | 1M | 0.50 | 2.50 | 0.625 | 0.05 |
| Sonnet 5.5 | 1M | 2.00 | 10.00 | 2.50 | 0.10 |
| Opus 5.5 | 1M | 4.00 | 20.00 | 5.00 | 0.20 |
For Haiku 5.5 with prompts exceeding 100K tokens, the input unit price of 0.50 is 1/4 of Sonnet 5.5's 2.00, 1/8 of Opus 5.5's 4.00, and half of Haiku 4.5's 1.00. The output unit price follows the same ratio: 2.50 is 1/4 of Sonnet 5.5's 10.00, 1/8 of Opus 5.5's 20.00, and half of Haiku 4.5's 5.00. For prompts of 100K tokens or less, the input unit price of 0.10 is 1/20 of Sonnet 5.5's price. Tasks that previously required Sonnet or Opus for 1M context become candidates for replacing with Haiku 5.5 from a pricing perspective.
Comparison with Non-Claude Models
GPT-6 Luna
The GPT-6 Luna Context window is 1,050,000 tokens. The GPT-6 Luna unit prices are from the Pricing section (Standard tier) of the Bedrock model card. The Haiku 5.5 unit prices are from the Global column in the previous section. The unit is USD / 1M tokens, and each cell shows "input / output" in that order.
| Model (Global) | Prompt 100K or less | Prompt over 100K up to 272K | Prompt over 272K |
|---|---|---|---|
| Haiku 5.5 | 0.10 / 0.50 | 0.50 / 2.50 | 0.50 / 2.50 |
| GPT-6 Luna (Global CRIS) | 0.10 / 0.50 | 0.10 / 0.50 | 0.20 / 0.75 |
The prices are equal for prompts of 100K tokens or less, and Luna has lower unit prices above 100K. The GPT-6 Luna model card states that exceeding 272K tokens applies the long-context price to the entire request. The unit prices for GPT-6 Luna Geo (US) and Mantle in-Region are 10% higher than Global.
Open Models
Unit prices are from the Bedrock Price List (us-east-1, Standard, published 2026-10-06T14:47:26Z), in USD / 1M tokens. Context length and availability start date are from each model card.
| Model | Input | Output | Context | Availability on Bedrock |
|---|---|---|---|---|
| gpt-oss-20b | 0.07 | 0.30 | 128K | August 2025 |
| gpt-oss-120b | 0.15 | 0.60 | 128K | August 2025 |
| Gemma 4 26B A4B | 0.13 | 0.40 | 256K | March 2026 |
| Gemma 4 31B | 0.14 | 0.40 | 256K | March 2026 |
| DeepSeek V3.2 | 0.62 | 1.85 | 164K | December 2025 |
| Kimi K3 | 3.30 | 16.50 | 1M | September 2026 |
Among the models in this table, only Kimi K3 supports a 1M context. Kimi K3's unit price is approximately 6.6 times that of Haiku 5.5 (Global) for prompts over 100K, making it more expensive than Sonnet 5.5.
Summary
Following Opus 5.5 and Sonnet 5.5 in September 2026, Claude Haiku 5.5 was released in October.
Claude Haiku 5.5 arrived 11 months after the release of Haiku 4.5. For prompts of 100K tokens or less, the unit price is 1/10 of Haiku 4.5, making it comparable to open models.
When prompts exceed 100K tokens the price increases fivefold, but even then, context up to 1M tokens can be handled at 1/4 the unit price of Sonnet 5.5.
Claude Haiku 5.5 comes with a JP inference profile available from the time of GA. It is easy to evaluate by calling Bedrock Runtime from the AWS CLI with IAM authentication, so please give it a try.
