What is Claude Haiku 5.5?
Claude Haiku 5.5 is Anthropic's small, fast model for high-volume and latency-sensitive work. Released October 7, 2026, it is available through the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS.
The model supports text and image input, text output, adaptive thinking and a one-million-token context window. Anthropic positions it for classification, extraction, routing, compaction, summarization, browser use and subagent work rather than the hardest long-horizon coding tasks reserved for Sonnet, Opus or Fable.
QUICK VERDICT: Claude Haiku 5.5 is Anthropic's fastest current model and one of the cheapest capable options for prompts up to 100,000 tokens. It gains a one-million-token context window, adaptive thinking and much stronger agent support, but its fivefold long-prompt price step makes prompt length a central architectural decision.
Field | Verified value |
|---|---|
Provider | Anthropic |
Release date | October 7, 2026 |
License | Proprietary |
Claude API ID | claude-haiku-5-5 |
Context window | 1 million tokens |
Maximum output | 128,000 tokens synchronously; 300,000 in supported Batch beta |
Modalities | Text and image input; text output |
Reliable knowledge cutoff | June 2026 |
Thinking | Adaptive; medium effort by default |
Comparative latency | Fastest in Anthropic's current lineup |
Base price up to 100K prompt | $0.10 input and $0.50 output per million tokens |
Last verified | October 22, 2026 |
What changed from Claude Haiku 4.5
Haiku 5.5 is a substantial capability and platform update over Haiku 4.5. It adds the newer tokenizer, adaptive thinking, a much larger context window, longer outputs and broader cloud availability. The main commercial change is tiered pricing based on whether a prompt exceeds 100,000 tokens.
Area | Claude Haiku 5.5 change |
|---|---|
Context | One million tokens with up to 128K synchronous output |
Thinking | Adaptive thinking controlled through the effort parameter |
Tokenizer | Newer tokenizer counts the same text as about 30% more tokens than Haiku 4.5 |
Pricing | Very low base rate up to 100K prompt tokens; fivefold higher rates above 100K |
Platforms | Claude API, AWS, Google Cloud and Microsoft Foundry availability |
Agent use | Improved fit for browser use, compaction and bounded subagent tasks |
A newer model is not automatically the right migration target. Compare Claude Haiku 5.5 with Claude Haiku 4.5 using the same prompts, tools, source material, effort controls and acceptance rubric. Record quality, latency, retries, review time and total cost per accepted result.
Capabilities and best use cases
Haiku 5.5 is optimized for tasks where speed and unit economics matter. Adaptive thinking lets it spend more effort when needed, but applications should set effort explicitly and measure the latency distribution. Anthropic advises omitting temperature, top_p and top_k for this model.
Capability | Practical use |
|---|---|
Classification | High-volume labels, triage and policy-based decisions |
Extraction | Structured fields from text, documents and images |
Routing | Choose tools, models or workflows and escalate uncertain cases |
Summarization | Compaction, notes and concise representations of larger context |
Subagents | Bounded parallel work inside a larger agent system |
Browser use | Fast tool-assisted navigation with application-level safeguards |
The strongest documented fit is high-volume, latency-sensitive classification, extraction, routing, summarization, compaction and subagent work. Reliability still depends on retrieval, tool permissions, prompt design, output validation and escalation policy. Test the complete system instead of evaluating the model name in isolation.
Benchmark evidence
Anthropic reports broad improvements over Haiku 4.5 in capability and alignment. It also distinguishes the intended workload clearly: Haiku 5.5 is strong for narrowly scoped high-volume tasks, while Sonnet 5.5 and Opus 5.5 remain better choices for difficult long-horizon coding. That boundary matters more than isolated launch-chart wins.
Evaluation area | Reported evidence | How to interpret it |
|---|---|---|
General capability | Provider reports major improvement over Haiku 4.5 | Validate on the same prompts and effort setting |
Agentic work | Designed for bounded subagent and browser tasks | Not positioned as the first choice for the hardest long-horizon coding |
Alignment | Fewer misaligned behaviors in Anthropic evaluations | See the system card for methods and limitations |
Misuse cooperation | Lower willingness to cooperate with misuse | Provider safety evaluation |
Latency | Fastest current Claude model | Actual latency depends on prompt, output and effort |
Benchmark results depend on the model version, effort, sampling, tools, scaffolding and dataset release. Do not combine scores from different harnesses into a synthetic ranking. Use public evidence to choose candidates, then test those candidates on frozen tasks from the intended workflow.
Pricing and access
Claude Haiku 5.5 is available on Anthropic's first-party API and major cloud platforms. The same dateless model ID is used on the Claude API, Google Cloud, Microsoft Foundry and Claude Platform on AWS; Amazon Bedrock uses the anthropic.claude-haiku-5-5 form.
Pricing changes sharply at 100,000 prompt tokens. For prompts at or below that threshold, input costs $0.10 and output $0.50 per million tokens. Above it, input costs $0.50 and output $2.50. Cache writes and reads follow the same fivefold step. Batch processing discounts input and output by 50%.
Usage or access item | Current value |
|---|---|
Input up to 100K prompt tokens | $0.10 per million tokens |
Output up to 100K prompt tokens | $0.50 per million tokens |
Input above 100K prompt tokens | $0.50 per million tokens |
Output above 100K prompt tokens | $2.50 per million tokens |
Cache read | $0.01 below threshold; $0.05 above threshold |
Batch API | 50% discount on input and output |
Token price is only part of deployment cost. Include prompt construction, cached-input behavior, tool calls, retries, wall time, failed-task recovery and human review. A cheaper model can be more expensive operationally when it needs repeated attempts or produces artifacts that require substantial correction.
Limitations and deployment risks
- The fivefold price step above 100,000 prompt tokens can materially change workload economics.
- The newer tokenizer counts the same text as roughly 30% more tokens than Haiku 4.5.
- Haiku is not Anthropic's recommended model for the hardest long-horizon coding tasks.
- Adaptive thinking can increase latency and cost if effort is not measured and controlled.
- A million-token window does not guarantee accurate retrieval or instruction retention throughout the prompt.
- Cybersecurity safeguards are more restrictive than Haiku 4.5 for some requests.
- Migration testing should cover prompts, output length, tokenizer effects, caching and tool behavior.
High-stakes legal, medical, financial, scientific and security work needs qualified review. Log the exact model version and settings, separate untrusted content from system instructions, minimize tool permissions and require confirmation before irreversible or externally visible actions.
How to evaluate Claude Haiku 5.5
Build a frozen evaluation set of 20 to 50 real tasks. Include routine work, difficult edge cases, adversarial inputs, long-context examples and cases where the correct behavior is to stop or escalate. Compare Claude Haiku 5.5 with Claude Haiku 4.5 and at least one neighboring model under equivalent conditions.
Test area | What to record |
|---|---|
Task completion | Pass or fail against a written acceptance rubric |
Reliability | Repeated-run success rate, variance and silent failures |
Factuality | Unsupported claims, source use, quotations and citation accuracy |
Tool use | Wrong calls, retries, recovery and permission-boundary failures |
Long context | Retrieval accuracy, instruction retention and cost at realistic lengths |
Efficiency | Wall time, input, output, cached tokens, tool charges and review time |
Safety | Prompt injection, sensitive-data handling and irreversible-action controls |
Choose the least expensive configuration that clears the acceptance threshold with a safety margin. Re-run the suite when the model, system prompt, effort, retrieval layer, tool definitions or approval policy changes.
Who should use it?
Situation | Recommendation |
|---|---|
Strong fit | high-volume, latency-sensitive classification, extraction, routing, summarization, compaction and subagent work |
Pilot first | Long-running agents, large contexts, computer use and workflows with several tools |
Escalate | Ambiguous or consequential work that does not reliably clear the evaluation threshold |
Avoid unsupervised use | Irreversible actions, sensitive data or high-stakes decisions without monitoring and approval |
A migration should be driven by measured outcomes. Keep Claude Haiku 4.5 available during the pilot, record where each model succeeds or fails, and use routing when different task classes have different quality and cost requirements.
Frequently asked questions
Is Claude Haiku 5.5 open source?
No. Claude Haiku 5.5 is a proprietary model controlled by Anthropic. Access, lifecycle and serving behavior depend on the provider and supported distribution platforms.
How much does Claude Haiku 5.5 cost?
$0.10 per million tokens. Review the full table because caching, prompt length, processing mode, platform and output volume can materially change the total.
What is Claude Haiku 5.5 best used for?
Its strongest documented fit is high-volume, latency-sensitive classification, extraction, routing, summarization, compaction and subagent work. Start with a supervised pilot and retain human sign-off for consequential work.
Should I migrate from Claude Haiku 4.5?
Only after a side-by-side evaluation. Measure accepted-task quality, total cost, latency, output style, tool reliability and migration engineering. Newer does not always mean better for a specific operational workflow.
How should benchmark evidence for Claude Haiku 5.5 be interpreted?
Use provider results to identify promising workloads, then reproduce the comparison with the exact configuration, tools and acceptance criteria that matter to the deployment. Do not infer a site ranking from benchmark claims gathered under a different harness.
Related model guides
Browse the Best AI Models directory.
Compare with the Claude Haiku 4.5 guide.
Compare with the Claude Sonnet 5.5 guide.
Compare with the Claude Opus 5.5 guide.
Official sources and update policy
Anthropic Claude Haiku 5.5 announcement.
Claude Haiku 5.5 model documentation.
Checked October 22, 2026. We update this guide when Anthropic changes the specification, pricing, access, safety documentation or lifecycle. Vendor benchmark claims are attributed and are not presented as independent testing. No vendor payment or affiliate relationship determined inclusion.