What Is Claude Haiku 5.5?

Claude Haiku 5.5 is Anthropic’s smallest and fastest current model, released on October 7, 2026, with the API ID claude-haiku-5-5. It costs $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100K tokens, and $0.50/$2.50 for prompts over 100K. In its launch post, Anthropic says it costs “around 75% less to run” than Haiku 4.5 on average and calls it “our fastest model to date.” It is also the first Haiku with an effort setting and a 1M-token context window.

Try Apidog today

This guide covers the specs, two-tier pricing, migration changes from Haiku 4.5, benchmarks, availability, and when to switch. For more pricing detail, see the Haiku 5.5 pricing breakdown. To call the model while you read, Apidog can send the Messages request, keep your key in an environment variable, and show the response alongside token usage.

Claude Haiku 5.5 specs at a glance

Spec	Claude Haiku 5.5
Release date	October 7, 2026
Model ID (Claude API, Google Cloud, Microsoft Foundry, Claude Platform on AWS)	claude-haiku-5-5 (fixed ID, no date suffix)
Model ID (Amazon Bedrock)	anthropic.claude-haiku-5-5
Context window	1M tokens (Haiku 4.5: 200K)
Max output	128K (Haiku 4.5: 64K); 300K on the Batch API with the output-300k-2026-03-24 beta header
Thinking	Adaptive only
Effort levels	low, medium (default), high, xhigh, max
Knowledge cutoff	June 2026
Price per MTok, prompts up to 100K tokens	$0.10 input, $0.50 output
Price per MTok, prompts over 100K tokens	$0.50 input, $2.50 output
Retirement	Not sooner than October 7, 2027

Where Haiku 5.5 sits in the Claude lineup

Model	Input / output per MTok	Default effort	Latency
Claude Fable 5.1	$10 / $50	high	Slower
Claude Opus 5.5	$4 / $20	medium	Moderate
Claude Sonnet 5.5	$2 / $10	high	Fast
Claude Haiku 5.5	$0.10 / $0.50 for prompts up to 100K tokens	medium	Fastest

All four share a 1M context window, 128K max output, and a June 2026 cutoff. Anthropic built Haiku 5.5 for high-volume, latency-sensitive work such as classification, extraction, routing, summaries, live support, browser use, and subagent tasks under Opus 5.5 or Sonnet 5.5.

The speed claim has an important caveat: it is the fastest model “at each model’s standard speed,” but slower than Opus in Fast Mode. Anthropic does not publish a tokens-per-second figure.

Claude Haiku 5.5 pricing

Pricing depends on prompt length. Once a prompt exceeds 100,000 tokens, every input and output token uses the higher rate.

Source: Claude API pricing.

The “around 75% less” figure is an average:

• Up to 100K tokens: 90% cheaper than Haiku 4.5.
• Over 100K tokens: 50% cheaper than Haiku 4.5.
• Anthropic says 90% of Haiku 4.5 requests are below the 100K-token threshold.

The comparison already accounts for the new tokenizer, which produces approximately 30% more tokens for the same text.

US-only inference (inference_geo) costs 1.1x for every token category. The minimum cacheable prompt is also lower: 512 tokens instead of 4,096 for Haiku 4.5.

For prompts under 100K tokens, Haiku 5.5 matches GPT-6 Luna’s listed price: $0.10 input, $0.50 output, $0.01 cache read, and $0.125 cache write per MTok. The thresholds differ: Luna’s surcharge begins above 272K input tokens. See the Haiku 5.5 vs GPT-6 Luna comparison and the pricing guide for worked cost examples.

What changed from Haiku 4.5

Haiku 5.5 increases the context window by 5x, doubles maximum output, and adds effort control. However, the migration guide warns that some Haiku 4.5 request formats return 400 errors.

Fix these breaking changes

1. Replace manual thinking budgets

This Haiku 4.5 configuration fails:

   {
     "thinking": {
       "type": "enabled",
       "budget_tokens": 2000
     }
   }

Use adaptive thinking with effort instead:

   {
     "thinking": {
       "type": "adaptive"
     },
     "output_config": {
       "effort": "medium"
     }
   }

1. Remove sampling parameters

Do not send temperature, top_p, or top_k.

A non-default temperature or top_p fails. Any top_k fails. Sending both temperature and top_p also fails.

1. Remove assistant prefills

Assistant-prefilled messages are rejected, including when thinking is disabled.

1. Update the computer-use tool

On the Claude API and Google Cloud, use:

   {
     "type": "computer_toolset_20260801"
   }

The older computer_20250124 tool fails.

1. Keep message history append-only

Do not modify system, tools, or earlier messages before a returned thinking block. Those edits fail validation.

Account for behavioral changes

These changes do not necessarily return errors, but they can change output, token usage, and application logic:

• Thinking display: Thinking blocks are empty by default except for a signature. Use "display": "summarized" to receive summaries.
• Token counts: The tokenizer can produce roughly 30% more tokens for the same text. max_tokens includes thinking tokens.
• Forced tool use: tool_choice with any or a named tool still works, but the response does not include a thinking block.
• Refusals: Safety classifiers can return stop_reason: "refusal" without a server-side fallback.
• Priority Tier: Haiku 5.5 does not support Priority Tier.

Thinking can be disabled at high effort or below. Disabling thinking at xhigh or max returns a 400 error.

Haiku 4.5 is not deprecated and remains active. For before-and-after payloads, see the Haiku 5.5 vs Haiku 4.5 guide.

Claude Haiku 5.5 benchmarks

Haiku 5.5 benchmark scores use max effort and are mostly averaged across five trials. Artificial Analysis ran GDPval-AA and AA-Briefcase, Cognition ran FrontierCode, and Anthropic graded Chartography, a Surge AI benchmark, with its own implementation.

Read several comparisons carefully:

• FrontierCode compares Haiku at max with Sonnet 5.5 at xhigh. At max, Sonnet scored 46.2%, below Haiku’s 46.4%.
• Anthropic ran Luna’s OSWorld score through OpenAI’s API.
• Luna’s Terminal-Bench score came from the public leaderboard using Codex CLI.
• At default medium effort, Haiku 5.5 scored 1277 on GDPval-AA and 1372 on AA-Briefcase.

Anthropic says Sonnet 5.5 and Opus 5.5 remain better choices for complex agentic coding tasks. No independent leaderboard had a Haiku 5.5 result as of October 8, 2026. Cursor reports 48.4% on its own CursorBench at max effort.

For per-effort scores and cost comparisons, see the benchmarks deep dive.

Where you can use Claude Haiku 5.5

Surface	Access	Notes
Claude API	claude-haiku-5-5	Pay per token
Amazon Bedrock	anthropic.claude-haiku-5-5	Billed through AWS Marketplace
Google Cloud, Microsoft Foundry, Claude Platform on AWS	claude-haiku-5-5	Same ID as the Claude API
Claude.ai (web, iOS, Android)	Model picker	Free, Pro, Max, Team, and Enterprise users can select it
Claude Code	/model claude-haiku-5-5	v2.1.293 or later; paid plans
Vercel AI Gateway	anthropic/claude-haiku-5.5	From $0.10 / $0.50
Cursor	claude-haiku-5-5	Other Models pool

As of October 8, 2026, Haiku 5.5 was not available on OpenRouter or announced for GitHub Copilot.

Chat access on the Free plan does not provide an API key. The API is pay-as-you-go beyond the small free credit Anthropic gives new users for testing. Max and Team plans include monthly API credits:

• Max 5x: $100
• Max 20x: $200
• Team: up to $500 pooled

These credits do not cover Claude Code. See the free access guide for the available free options.

In Claude Code, the haiku alias means Haiku 5.5 only when using the Anthropic API. Elsewhere, it still means Haiku 4.5. See Haiku 5.5 in Claude Code for subagent setups.

Who should switch to Haiku 5.5

• You run Haiku 4.5: Switch after fixing the five breaking changes. Haiku 5.5 beats Haiku 4.5 on every launch-table row where both models have a score, at a lower listed price.
• You use Sonnet or Opus for simple repeated tasks: Test Haiku 5.5 for classification, extraction, summaries, routing, and subagent work. Keep larger models for complex agentic coding.
• Your prompts often exceed 100K tokens: Model the two pricing tiers before migrating. Above the threshold, input tokens cost 5x more.
• You do security work: Haiku 5.5’s cyber safeguards still block penetration testing.

In Anthropic’s launch post, Asana reports more than 30% lower latency on task completions, while Box reports a score 11 points above Haiku 4.5 at about half the latency. These are customer claims, not independent tests.

The system card notes that Haiku 5.5 over-refused more than any other model in Anthropic’s automated audit. Measure refusal rates against your own production prompts.

Send your first Haiku 5.5 request

Start with a minimal Messages API request:

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-haiku-5-5",
    "max_tokens": 4000,
    "thinking": {"type": "adaptive", "display": "summarized"},
    "output_config": {"effort": "low"},
    "messages": [
      {
        "role": "user",
        "content": "Classify this support ticket as billing, bug, or feature request: The export button returns a 500 error."
      }
    ]
  }'

Implementation checklist:

1. Do not send sampling parameters or assistant prefills.
2. Parse response content blocks by type, because the response can start with a thinking block.
3. Check stop_reason and handle refusal and max_tokens.
4. Run the same workload with low and medium effort.
5. Compare usage across effort levels before setting a production default.

In Apidog, import the cURL command as a new request and store the key in an ANTHROPIC_API_KEY environment variable. Add assertions that stop_reason is not refusal or max_tokens, then duplicate the request to compare effort settings and token usage.

For caching, batch processing, and streaming, see the Haiku 5.5 API guide. If you need credentials, use the Anthropic API key guide.

FAQ

When was Claude Haiku 5.5 released?

October 7, 2026, nine days after Sonnet 5.5.

Is Claude Haiku 5.5 free?

In the Claude.ai chat app, Free users can select it. The API is paid except for small test credits for new users. See every free route.

What is the Claude Haiku 5.5 context window?

1M tokens, with 128K maximum output on the Messages API.

Is Haiku 4.5 being retired?

Not yet. It remains listed as active, with retirement not sooner than October 15, 2026.

Next step

Pick one high-volume task you run today, such as ticket triage or document summarization. Send it to claude-haiku-5-5 at low and medium effort, validate the output, and compare usage against your current model.

Download Apidog to keep requests and assertions in one project, then use the pricing guide to turn token counts into a monthly cost estimate.