Claude Haiku 5.5 is Anthropic's small, high-volume Claude model, announced on October 7, 2026. On the Claude Platform the model ID is claude-haiku-5-5. Official list prices start at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, then jump to $0.50 / $2.50 when the prompt is longer. That tiered schedule is the main difference from Haiku 4.5 and Sonnet 5.5, which use flat per-token rates.
Anthropic positions Haiku 5.5 for classification, extraction, routing, summarization, compaction, database-style lookups, live customer support, browser use, and subagent work beside Sonnet 5.5 or Opus 5.5. It is the first Haiku-class model with an adjustable effort setting for adaptive thinking.
Official API pricing
USD per million tokens from Anthropic's Haiku 5.5 announcement, the model overview, and the pricing page, checked October 29, 2026:
| Item | Haiku 5.5 (prompt ≤100k) | Haiku 5.5 (prompt >100k) | Haiku 4.5 | Sonnet 5.5 |
|---|---|---|---|---|
| Input | $0.10 | $0.50 | $1.00 | $2.00 |
| Output | $0.50 | $2.50 | $5.00 | $10.00 |
| Cache read | $0.01 | $0.05 | $0.10 | $0.10 |
| 5-minute cache write | $0.125 | $0.625 | $1.25 | $2.50 |
| 1-hour cache write | $0.20 | $1.00 | $2.00 | $4.00 |
Prompt length for the Haiku 5.5 threshold counts all input tokens on that request, including cache reads and writes. Batch API requests take 50% off input and output on both Haiku 5.5 tiers. Anthropic's footnote says Haiku 5.5 is priced 90% lower than Haiku 4.5 for requests up to 100k tokens and 50% lower above that; across Haiku 4.5 traffic where about 90% of requests were in the shorter bucket, average run cost falls around 75%, after also accounting for the newer tokenizer.
Sonnet 5.5 cache reads were cut to $0.10 (from $0.20) in the same launch window. That is a Sonnet change, not a Haiku 5.5 rate.
For a dedicated rate walkthrough, see Claude Haiku 5.5 API Pricing.
Specs and model IDs
From the Haiku 5.5 overview:
| Spec | Value |
|---|---|
| Claude API model ID | claude-haiku-5-5 |
| Amazon Bedrock | anthropic.claude-haiku-5-5 |
| Google Cloud / Microsoft Foundry / Claude Platform on AWS | claude-haiku-5-5 |
| Context window | 1M tokens |
| Max output | 128K tokens (up to 300K on Message Batches with beta header output-300k-2026-03-24) |
| Thinking | Adaptive; default effort medium |
| Comparative latency | Fastest in Anthropic's current lineup table |
| Input → output | Text and images → text |
| Knowledge / training cutoff | June 2026 |
| Status | Active (latest); retirement not sooner than October 7, 2027 |
Haiku 5.5 uses the newer tokenizer shared with Claude 4.7 and later models. The same text counts as about 30% more tokens than on Haiku 4.5. Thinking blocks from Haiku 5.5 work only in the account that produced them, or in a linked account.
Where Haiku 5.5 fits versus Haiku 4.5 and Sonnet 5.5
Anthropic's launch page and overview draw a clear split:
- Versus Haiku 4.5. Haiku 5.5 is cheaper on the published schedule, adds adaptive effort, and posts large gains on Anthropic's published suites (for example GDPval-AA v2.1 at 1620 vs 735, Terminal-Bench 4.0 at 39.2% vs 0.0%, OSWorld 2.1 offline subset at 72.4% vs 15.7%). Those numbers are Anthropic's evaluation table, not an independent re-run.
- Versus Sonnet 5.5. Sonnet stays the better pick for complex agentic coding (Terminal-Bench 4.0: Sonnet 5.5 70.6% vs Haiku 5.5 39.2% on the same announcement table). Haiku 5.5 is aimed at narrowly scoped, high-volume work that used to be too expensive at prior Haiku rates—compaction, summarization, routing, and subagent pulls while a larger model holds the main thread.
- Effort knob. Haiku 5.5 is the first Haiku with adjustable effort. Anthropic publishes accuracy-vs-cost charts at each effort setting for OSWorld 2.1, GDPval-AA v2.1, and Humanity's Last Exam; pick effort from cost and latency needs, not from a single score.
Customer quotes on the launch page (Asana, HubSpot, AlphaSense, Box, Gamma, Cognition) describe lower latency and better scores on their internal suites. Treat them as attributed early testing, not as AI Rank measurements.
Safety notes worth knowing
Anthropic reports stronger alignment results than Haiku 4.5 on its evaluations, with fewer misaligned behaviors and less willingness to cooperate with misuse. Cybersecurity safeguards are tighter than Haiku 4.5 but somewhat looser than Sonnet 5.5: more defensive cyber work is allowed, while penetration testing and attacker-leaning techniques stay blocked. Biology safeguards match Sonnet 5 / Sonnet 5.5 / Opus 5: research questions are allowed; harmful requests are restricted. Broader biology or cyber programs go through Anthropic's Life Sciences and Cyber verification programs. Details sit in the Haiku 5.5 System Card linked from the overview.
How to choose in practice
- Short, repeated work under 100k prompt tokens — Haiku 5.5's $0.10 / $0.50 tier is the published bargain in this family.
- Long prompts over 100k — budget the $0.50 / $2.50 tier; it is still half of Haiku 4.5's flat $1 / $5, but no longer the extreme discount.
- Hard agentic coding or long multi-step planning — Sonnet 5.5 or Opus 5.5 remain the models Anthropic points to for Terminal-Bench-class work.
- Tokenizer surprise — expect roughly 30% more tokens for the same text than Haiku 4.5; re-estimate cost with the new tokenizer, not old prompt lengths alone.
Related pages on AI Rank: Haiku 5.5 API Pricing, Haiku 5.5 Benchmarks, Sonnet 5.5, Claude Opus 5.5.
Sources
- Anthropic: Introducing Claude Haiku 5.5 (October 7, 2026) — positioning, pricing table, benchmarks, availability
- Claude Platform: Claude Haiku 5.5 overview — model IDs, specs, effort, cutoff, retirement floor
- Claude Platform: Pricing — Haiku 5.5 two-tier rates, cache, batch, Sonnet 5.5 cache read
Keep exploring: compare current AI model rankings on AI Rank.
Prepared with automated editorial assistance and checked against Anthropic's Haiku 5.5 product page, model overview, and pricing docs on October 29, 2026 (America/Los_Angeles). We did not run our own benchmarks for this article.