Claude Haiku is Anthropic's fastest and lowest-priced Claude model line, whose first version, Claude 3 Haiku, carries a March 2024 snapshot date, and four Haiku versions have appeared in Anthropic's documentation since then. This page compares every Haiku version side by side: API IDs, release dates, status, context window, maximum output, prices, documented behaviour changes, published benchmark figures and retirement dates. It is written for developers migrating between versions, teams running high-volume workloads such as classification and routing, and buyers estimating cost. Versions are listed by release date, newest first, an objective order that implies no ranking. Every fact comes from Anthropic's own documentation and announcements, cited in the sources list.
Last checked: 2026-11-15.
This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider. It uses Anthropic's public documentation as of November 2026. Model details change, so confirm against Anthropic's current documentation.
Key takeaways
- Two Haiku versions are active. Anthropic's deprecations page lists claude-haiku-5-5 and claude-haiku-4-5-20251001 as Active; Haiku 3.5 and Haiku 3 are retired on the Claude API.
- Haiku 4.5's retirement floor has passed. Its "not sooner than" date was 2026-10-15; as of 2026-11-15 it is still Active with no deprecation announced, so watch the deprecations page.
- Haiku 5.5 has tiered pricing. It costs $0.10 input and $0.50 output per million tokens for prompts up to 100K tokens, and $0.50 and $2.50 above that; Haiku 4.5 is $1 and $5.
- Context grew from 200K to 1M tokens. Haiku 4.5 lists 200K tokens and 64K output; Haiku 5.5 lists 1M tokens and 128K output.
- Haiku 5.5 can break Haiku 4.5 code. Anthropic states manual extended thinking, sampling parameters and assistant prefill can return 400 errors on Haiku 5.5.
What is Claude Haiku?
Claude Haiku is one of five Claude model lines in Anthropic's documentation, alongside Fable, Mythos, Opus and Sonnet. Anthropic's models overview describes Haiku 5.5 as "for high-volume, latency-sensitive tasks such as classification, extraction, and routing", rates its latency as Fastest and lists it with the lowest starting price of any current Claude model. Every Haiku model is proprietary: Anthropic publishes no weights and states no parameter count for any version. Haiku 5.5 uses an undated API identifier, claude-haiku-5-5; earlier versions carry a date snapshot, such as claude-haiku-4-5-20251001.
Anthropic's numbering is not continuous across the line. There was no Haiku 4, 4.6 or 5; the line moved from Haiku 3.5 (November 2024) to Haiku 4.5 (October 2025) and then to Haiku 5.5 (October 2026). For the wider Claude line-up, see the Anthropic Models hub.
Every model at a glance
As of 2026-11-15. Source: Anthropic models overview, model deprecations page and release notes.
| Model | API ID | Released | Status |
|---|---|---|---|
| Claude Haiku 5.5 | claude-haiku-5-5 | 2026-10-07 | Active |
| Claude Haiku 4.5 | claude-haiku-4-5-20251001 (alias claude-haiku-4-5) | 2025-10-15 | Active |
| Claude Haiku 3.5 | claude-3-5-haiku-20241022 | 2024-11-04 | Retired (Claude API) |
| Claude Haiku 3 | claude-3-haiku-20240307 | Not stated in the release notes checked (snapshot date 2024-03-07) | Retired |
Haiku 5.5 and 4.5 take text and image input and return text, per Anthropic's models overview. Haiku 3.5 launched as a text-only model and gained vision support on 2025-02-24. Open weights: none for any version. Licence: proprietary. Parameters: not stated by Anthropic.
When was each model released?
As of 2026-11-15. Dates are as shown in Anthropic's Claude Platform release notes.
| Date | Event | API ID |
|---|---|---|
| 2024-11-04 | Claude Haiku 3.5 available on the Claude API as a text-only model | claude-3-5-haiku-20241022 |
| 2025-02-24 | Vision support added to Haiku 3.5 | claude-3-5-haiku-20241022 |
| 2025-10-15 | Claude Haiku 4.5 released for real-time, high-volume and cost-sensitive uses | claude-haiku-4-5-20251001 |
| 2025-12-04 | Structured outputs supported on Haiku 4.5 | claude-haiku-4-5-20251001 |
| 2025-12-19 | Haiku 3.5 deprecation announced | claude-3-5-haiku-20241022 |
| 2026-01-29 | Structured outputs generally available on Haiku 4.5 | claude-haiku-4-5-20251001 |
| 2026-02-19 | Haiku 3.5 retired on the Claude API; Haiku 3 deprecation announced | claude-3-5-haiku-20241022, claude-3-haiku-20240307 |
| 2026-04-16 | Haiku 4.5 available self-serve in Amazon Bedrock | claude-haiku-4-5-20251001 |
| 2026-04-20 | Claude Haiku 3 retired | claude-3-haiku-20240307 |
| 2026-10-07 | Claude Haiku 5.5 released | claude-haiku-5-5 |
How much does each model cost?
As of 2026-11-15. USD per million tokens, from Anthropic's pricing page. Retired versions are no longer listed there and are omitted.
| Model | Input | Cached input | Cache write | Output | Batch (input / output) | Notes stated by provider |
|---|---|---|---|---|---|---|
| Claude Haiku 5.5 (prompts up to 100K tokens) | $0.10 | $0.01 | $0.125 (5m), $0.20 (1h) | $0.50 | $0.05 / $0.25 | Tiered by prompt length |
| Claude Haiku 5.5 (prompts over 100K tokens) | $0.50 | $0.05 | $0.625 (5m), $1.00 (1h) | $2.50 | $0.25 / $1.25 | Whole request priced at the higher tier |
| Claude Haiku 4.5 | $1.00 | $0.10 | $1.25 (5m), $2.00 (1h) | $5.00 | $0.50 / $2.50 | Flat pricing |
How the tier works: Anthropic's pricing page states that prompt length for the 100K threshold includes cache reads and cache writes, and each request is priced on its own, so a request over the threshold pays the higher rate even if part of the prompt is a cache hit. Haiku 5.5 is the only Sonnet or Haiku model with tiered pricing.
Tokeniser: Anthropic's Haiku 5.5 announcement states it uses an updated tokeniser that uses slightly more tokens per task, and the pricing page states Claude 4.7 and later models produce about 30% more tokens for the same text than the previous tokeniser. Re-measure token counts on your own prompts before comparing Haiku 5.5 cost with Haiku 4.5.
Data residency: setting inference_geo to "us" applies a 1.1x multiplier, which the pricing page states applies to Haiku 5.5's over-100K tier. Fast mode is not offered for any Haiku version on the pricing page. For cost guardrails, see How to set spend limits on LLM API keys.
What changed between models?
Only changes Anthropic documents in its release notes and announcements are listed.
- Haiku 3 to 3.5: Haiku 3.5 launched on the Claude API as a text-only model on 2024-11-04 and gained vision support on 2025-02-24.
- Haiku 3.5 to 4.5: Anthropic's announcement describes Haiku 4.5 as a drop-in replacement for Haiku 3.5, states coding performance similar to Sonnet 4 at one-third the cost and more than twice the speed, and prices it at $1/$5. The models overview notes that Haiku 4.5's Models API entry reports server_tools.code_execution.supported as true but top-level code_execution.supported as false. Structured outputs reached general availability on 2026-01-29.
- Haiku 4.5 to 5.5: manual extended thinking, sampling parameters and assistant prefill can return 400 errors. The announcement describes Haiku 5.5 as the first Haiku-class model with an adjustable effort setting, with an updated tokeniser and cybersecurity safeguards more restrictive than Haiku 4.5's but less restrictive than Sonnet 5.5's. The models overview lists adaptive thinking with a default effort of medium. Anthropic states Haiku 5.5 costs around 75% less to run on average than Haiku 4.5.
For background on thinking modes and effort settings, see Reasoning Models Explained.
What workloads does Anthropic position each model for?
This section records the provider's own positioning wording; it is not an assessment.
| Model | Positioning stated by provider | Related notes stated by provider |
|---|---|---|
| Claude Haiku 5.5 | High-volume, narrowly scoped work such as summarisation, compaction, classification and subagent tasks; Sonnet 5.5 and Opus 5.5 recommended for complex agentic coding | Described as Anthropic's fastest and most capable small model; computer use and browser use support added in beta to the Python and TypeScript SDKs; monthly API credits announced for Max, Ultra and Team subscribers |
| Claude Haiku 4.5 | Real-time, high-volume and cost-sensitive uses (release notes) | Coding performance described as similar to Sonnet 4 at one-third the cost; stated to exceed Sonnet 4 on some tasks such as computer use |
| Claude Haiku 3.5 | Not stated on the pages checked (2026-11-15) | Launched text-only; vision added later |
| Claude Haiku 3 | Not stated on the pages checked (2026-11-15) | Retired 2026-04-20 |
Worked cost example
Using the list prices above, a workload of 10,000 requests a day, each with a 20K-token prompt and a 1K-token response, uses 200 million input and 10 million output tokens a day. On Haiku 4.5 that is $200 input plus $50 output, $250 a day. On Haiku 5.5, every request sits in the lower tier, so it is $20 input plus $5 output, $25 a day, before any tokeniser difference. If the same workload used 120K-token prompts, every Haiku 5.5 request would move to the upper tier at $0.50/$2.50, so plan prompt size as carefully as model choice. Re-run the arithmetic with token counts measured on your own data, because the updated tokeniser changes the counts.
What is the context window of each model?
As of 2026-11-15. Context, maximum output and knowledge cutoff from Anthropic's models overview, deprecations page and release notes; reasoning controls as documented in the release notes, announcements and models overview.
| Model | Context window | Max output | Knowledge cutoff | Input | Output | Reasoning controls |
|---|---|---|---|---|---|---|
| Claude Haiku 5.5 | 1M tokens | 128K tokens | Jun 2026 (reliable) | Text, image | Text | Adaptive thinking, default effort medium; first Haiku with an adjustable effort setting; manual extended thinking can return a 400 error |
| Claude Haiku 4.5 | 200K tokens | 64K tokens | Not stated by Anthropic on the pages checked (2026-11-15) | Text, image | Text | Extended thinking (a 32K thinking budget is cited in the announcement) |
| Claude Haiku 3.5 | Not stated by Anthropic on the pages checked (2026-11-15) | Not stated by Anthropic on the pages checked (2026-11-15) | Not stated by Anthropic on the pages checked (2026-11-15) | Text at launch; image from 2025-02-24 | Text | Not stated by Anthropic on the pages checked (2026-11-15) |
| Claude Haiku 3 | Not stated by Anthropic on the pages checked (2026-11-15) | Not stated by Anthropic on the pages checked (2026-11-15) | Not stated by Anthropic on the pages checked (2026-11-15) | Not stated by Anthropic on the pages checked (2026-11-15) | Not stated by Anthropic on the pages checked (2026-11-15) | Not stated by Anthropic on the pages checked (2026-11-15) |
Haiku 5.5 lists a 1M-token context window and 128K maximum output, with the price tier changing at 100K prompt tokens. Haiku 4.5 lists 200K tokens and 64K output at flat pricing. Context windows for Haiku 3.5 and Haiku 3 are not stated on the current pages checked. Teams moving long-document workloads from Haiku 4.5 to Haiku 5.5 should model how many requests cross the 100K threshold, since those requests pay five times the lower-tier rate.
Capabilities and tools compared
As of 2026-11-15. Anthropic's models overview states that "all current models support text and image input, text output, multilingual capabilities, vision, and tool use". Cloud IDs are from the individual model overview pages on platform.claude.com.
| Model | Function calling and structured outputs | Computer use tool | Fine-tuning | Cloud IDs stated by provider |
|---|---|---|---|---|
| Claude Haiku 5.5 | Tool use supported (models overview) | Computer use and browser use support added in beta to the Python and TypeScript SDKs (announcement) | Not stated by Anthropic on the pages checked (2026-11-15) | Amazon Bedrock anthropic.claude-haiku-5-5; Google Cloud claude-haiku-5-5; Microsoft Foundry claude-haiku-5-5 |
| Claude Haiku 4.5 | Tool use supported (models overview); structured outputs supported from 2025-12-04, generally available from 2026-01-29 | Stated to exceed Sonnet 4 on some tasks such as computer use (announcement) | Not stated by Anthropic on the pages checked (2026-11-15) | Not stated by Anthropic on the pages checked (2026-11-15) |
| Claude Haiku 3.5 | Not stated by Anthropic on the pages checked (2026-11-15) | Not stated by Anthropic on the pages checked (2026-11-15) | Not stated by Anthropic on the pages checked (2026-11-15) | Not stated by Anthropic on the pages checked (2026-11-15) |
| Claude Haiku 3 | Not stated by Anthropic on the pages checked (2026-11-15) | Not stated by Anthropic on the pages checked (2026-11-15) | Not stated by Anthropic on the pages checked (2026-11-15) | Not stated by Anthropic on the pages checked (2026-11-15) |
Fine-tuning is not mentioned for any Haiku version on the Anthropic pages checked; partner clouds may publish their own options, so confirm in each cloud's catalogue.
Where is each model available?
| Model | Availability stated by provider |
|---|---|
| Claude Haiku 5.5 | "All platforms", including Amazon Web Services, Google Cloud and Microsoft Azure (announcement) |
| Claude Haiku 4.5 | Claude Code, Claude apps, Claude API, Amazon Bedrock and Google Cloud Vertex AI (announcement); self-serve in Amazon Bedrock from 2026-04-16 |
| Claude Haiku 3.5 | Retired on the Claude API on 2026-02-19; the release notes state it remained on Amazon Bedrock and Google Cloud at that date |
| Claude Haiku 3 | Retired on the Claude API |
Partner clouds publish their own model IDs and retirement schedules, which can differ from the Claude API. Confirm in each cloud's model catalogue.
Model cards and system cards
| Model | Model card or page | System card | Technical report |
|---|---|---|---|
| Claude Haiku 5.5 | Announcement | System card | Not stated by Anthropic on the pages checked (2026-11-15) |
| Claude Haiku 4.5 | Announcement | System card | Not stated by Anthropic on the pages checked (2026-11-15) |
System card links are as given in each announcement; this page does not summarise their contents.
Benchmarks
As of 2026-11-15. Figures are provider-reported, as stated in the text of each Anthropic announcement. Benchmark suites and versions differ between announcements, so figures in different rows are not directly comparable unless the same announcement reports both.
| Model | Provider-reported benchmarks | Independent leaderboard results |
|---|---|---|
| Claude Haiku 5.5 | OSWorld 2.1 (offline subset) 72.4%; Terminal-Bench 4.0 39.2%; Humanity's Last Exam 45.9% (no tools), 57.4% (with tools); FrontierCode 1.1 (Main) 46.4% (announcement, 2026-10-07) | Not checked for this page |
| Claude Haiku 4.5 | SWE-bench Verified 73.3% (50 trials); Terminal-Bench 40.21% without thinking, 41.75% with a 32K thinking budget (announcement, 2025-10-15). The Haiku 5.5 announcement reports Haiku 4.5 at 15.7% on OSWorld 2.1 (offline subset) and 0.0% on Terminal-Bench 4.0 | Not checked for this page |
Deprecations and retirements
As of 2026-11-15. Source: Anthropic model deprecations page. Anthropic states it gives at least 60 days' notice before retiring publicly released models.
| Model | API ID | Notice or deprecation date | Retirement or shutdown date | Stated replacement |
|---|---|---|---|---|
| Claude Haiku 5.5 | claude-haiku-5-5 | Active | Not sooner than 2027-10-07 | None stated |
| Claude Haiku 4.5 | claude-haiku-4-5-20251001 | Active | Not sooner than 2026-10-15 | None stated |
| Claude Haiku 3.5 | claude-3-5-haiku-20241022 | 2025-12-19 | 2026-02-19 (Claude API) | claude-haiku-4-5-20251001 |
| Claude Haiku 3 | claude-3-haiku-20240307 | 2026-02-19 | 2026-04-20 | claude-haiku-4-5-20251001 |
Requests to a retired ID return an error. Because Haiku 4.5's "not sooner than" date has passed, a deprecation notice with at least 60 days' warning could be published at any time. For lifecycle planning, see the AI Model Retirement Guide and the LLM Deprecation and Retirement Tracker.
Data handling and security
Anthropic's zero data retention page designates only the Fable and Mythos 5.x models as Covered Models requiring 30-day retention; no Haiku version is on that list. Regardless of arrangement, the page states flagged content may be retained for up to 2 years. Confirm retention terms in your own agreement. Haiku 5.5's cybersecurity safeguards are stricter than Haiku 4.5's, which can change how security tooling built on Haiku behaves after an upgrade.
Haiku is often used for subagents and high-volume pipelines, which multiplies the number of services holding Claude API keys. See the Anthropic API Keys Security Guide, the LLMjacking Guide and the Claude Agent SDK Security Guide.
How to choose a model: factual criteria
| Criterion | What the documentation shows |
|---|---|
| Lifecycle runway | Haiku 5.5 retires not sooner than 2027-10-07; Haiku 4.5's floor (2026-10-15) has passed |
| List price per token | Haiku 5.5 from $0.10/$0.50; Haiku 4.5 $1/$5 |
| Prompt length pattern | Haiku 5.5 prompts over 100K tokens pay $0.50/$2.50; Haiku 4.5 is flat up to 200K |
| Context above 200K tokens | Haiku 5.5 only |
| Assistant prefill and sampling parameters | Can return 400 errors on Haiku 5.5 |
| Effort control | Haiku 5.5 is the first Haiku with an adjustable effort setting (default medium) |
Run your own evaluation on representative prompts before switching versions; provider benchmarks use different suites per release.
Practitioner checklist
- Inventory every service still calling claude-haiku-4-5-20251001 and plan its migration before a deprecation notice arrives.
- Pin production code to an explicit Haiku API ID and record it in the AI inventory.
- Test Haiku 5.5 migrations for 400 errors from extended thinking, sampling parameters and assistant prefill.
- Measure what share of requests exceed 100K prompt tokens before estimating Haiku 5.5 cost.
- Re-measure token counts on real prompts with the updated tokeniser.
- Check partner cloud catalogues separately, including any Haiku 3.5 workloads still on a cloud platform.
- Store Claude API keys in a secrets manager and set workspace spend limits.
Standards and references
- Anthropic: Model deprecations (accessed 2026-11-15)
- Anthropic: Models overview (accessed 2026-11-15)
- Anthropic: Pricing (accessed 2026-11-15)
- Anthropic: Claude Platform release notes (accessed 2026-11-15)
- Anthropic: Zero data retention (accessed 2026-11-15)
- Anthropic: Introducing Claude Haiku 5.5 (accessed 2026-11-15)
- Anthropic: Introducing Claude Haiku 4.5 (accessed 2026-11-15)
Related NHI Mgmt Group resources: Anthropic Models, Claude Sonnet Versions Compared, Small and Low-Cost LLMs Compared, LLM Pricing Tracker, LLM Security Guide.