As of November 10, 2026, Anthropic identifies Claude Haiku 5.5 as its latest Haiku model. For the direct Claude API, its model ID is claude-haiku-5-5. Anthropic positions it for high-volume, latency-sensitive work such as classification, extraction, and routing. The published context window is 1 million tokens, with up to 128,000 output tokens. (platform.claude.com)
Haiku is a candidate for frequent, focused tasks where response time and token cost matter. Whether it meets your quality and latency targets depends on your workload, so evaluate it on representative inputs before routing production traffic. If you use Haiku 4.5, Anthropic says it remains available and recommends considering migration to Haiku 5.5. (platform.claude.com)
Which Claude Haiku model should you use?
Anthropic’s model documentation labels Haiku 5.5 the latest Haiku model. The direct Claude API ID is claude-haiku-5-5; the Amazon Bedrock model ID is different: anthropic.claude-haiku-5-5. Access, billing, and regional availability depend on the provider and deployment route. (platform.claude.com)
Anthropic lists Haiku 4.5 as still available and recommends considering migration to Haiku 5.5. The available Haiku 4.5 page supports that legacy-status guidance, but this guide does not list its API ID or limits. Anthropic’s Haiku 3.5 system-prompt page concerns dated consumer-app prompts, not a current API model catalog. (platform.claude.com)
| Model | Direct Claude API ID | Published context / max output | Practical note |
|---|---|---|---|
| Claude Haiku 5.5 | claude-haiku-5-5 |
1M / 128K tokens | Latest Haiku model; positioned for high-volume, latency-sensitive tasks. |
| Claude Haiku 4.5 | Not listed here | Not listed here | Still available; Anthropic recommends considering migration to 5.5. |
Anthropic says Haiku 5.5 uses the same newer tokenizer as Claude 4.7 and later models, so the same text counts as approximately 30% more tokens than it does with Haiku 4.5. Recalculate token use and cost when migrating. A large context window is a capacity limit, not a promise that every long prompt will be fast or inexpensive. (platform.claude.com)
What is Claude Haiku suited for?
Anthropic positions Haiku 5.5 for high-volume, latency-sensitive work such as classification, routing, extraction, and subagent tasks. Those are intended use cases, not a guarantee that the model will meet your application’s accuracy or latency targets. (platform.claude.com)
A practical selection test is to compare models on representative inputs using task success, correction or escalation rate, end-to-end latency, and total cost. For example, a support-ticket classifier might be assessed by category accuracy and the share of cases needing human correction—not just whether an answer sounds plausible. This is an evaluation suggestion, not a reported test result.
Calling Haiku 5.5 through the Claude API
This Python example follows Anthropic’s documented SDK pattern for the Messages API. It is a minimal example for adaptation, not a claim that this code was run. Anthropic documents installing the anthropic package, supplying ANTHROPIC_API_KEY through the environment, creating a Messages request with model, max_tokens, and messages, and reading text blocks from the response. (platform.claude.com)
Install the SDK and set your API key in the environment:
pip install anthropic
export ANTHROPIC_API_KEY="your-api-key"
Then send a request and print any text blocks in the response:
from anthropic import Anthropic
client = Anthropic() # Uses ANTHROPIC_API_KEY from the environment.
message = client.messages.create(
model="claude-haiku-5-5",
max_tokens=256,
messages=[
{
"role": "user",
"content": "Classify this ticket as billing, technical, or account: My invoice shows an unexpected charge.",
}
],
)
for block in message.content:
if block.type == "text":
print(block.text)
The example’s prompt asks for a ticket category; the returned text depends on the request and is not shown as a tested result. For production, protect credentials, handle API errors and rate limits, monitor token usage, and evaluate model outputs against representative inputs.
Amazon Bedrock is a separate deployment and billing route. AWS lists anthropic.claude-haiku-5-5 as its Bedrock model ID and documents provider-specific endpoints, supported APIs, features, pricing, and regional availability. Confirm access and configuration for your intended AWS Region; do not assume that direct Claude API credentials, billing, or configuration apply to Bedrock. (docs.aws.amazon.com)
Claude Haiku 5.5 API pricing
Anthropic’s listed direct API rates are in USD per million tokens (MTok) and depend on prompt length: (platform.claude.com)
| Prompt length | Input | Output |
|---|---|---|
| Up to 100,000 tokens | $0.10 / MTok | $0.50 / MTok |
| Over 100,000 tokens | $0.50 / MTok | $2.50 / MTok |
These are Claude Platform rates, not an Amazon Bedrock price quote. Prompt caching and Batch API processing have separate pricing rules, so include them in an estimate if your application uses them. (platform.claude.com)
Example estimate: if a workload processes 1 million input tokens and generates 100,000 output tokens, with every request in the up-to-100,000-token prompt tier and no caching or batch discount, the direct API token charges would be about $0.15: $0.10 for input plus $0.05 for output. This is a rate calculation, not a measured bill; actual usage, prompt lengths, caching, and other applicable charges can change the total.
For Bedrock, check AWS’s current listing for your model, Region, and chosen pricing option rather than applying Anthropic’s direct API rates. (docs.aws.amazon.com)
Bottom line
Shortlist Claude Haiku 5.5 for frequent, focused work where latency and token cost are important, then validate quality and cost on your own workload. Use claude-haiku-5-5 on the direct Claude API. For Bedrock or another platform, confirm its provider-specific model ID, access, regional availability, and billing. When migrating from Haiku 4.5, recalculate token use because Anthropic reports a tokenizer-related increase. (platform.claude.com)
Sources
Claude Haiku 5.5 — Claude Platform Docs
Claude Haiku 4.5 — Claude Platform Docs
Python SDK — Claude Platform Docs
Get started with Claude — Claude Platform Docs