The short version
- One tenth of Haiku 4.5 on input and output. $0.10 / $0.50 up to 100,000 tokens, against $1.00 / $5.00 on Haiku 4.5.
- Above 100,000 tokens the rates jump 5x to $0.50 input / $2.50 output, so the official table reads “From $0.10 / $0.50”. The 1M window is real, but the price of filling it is not the headline rate.
- Cache reads are 10% of input: $0.01 (and $0.05 above 100k). Cache writes are $0.125 for the 5-minute TTL and $0.20 for one hour.
- Thinking is adaptive and on by default, with a default effort of
medium— more conservative than Sonnet 5.5'shigh. - The tokenizer changed. Verbatim: it “uses the same newer tokenizer as Claude 4.7 and later models, so the same text counts as approximately 30% more tokens than on Claude Haiku 4.5.” Compare cost with token counts in hand.
temperature,top_pandtop_kcan no longer be set to non-default values — they return a 400 error.
Pricing (USD per 1M tokens, from the official docs)
Values as published on the Haiku 5.5 overview page and the official pricing documentation; the Haiku 4.5 column matches our pricing comparison.
| Item | Haiku 5.5 (≤100k tokens) | Haiku 5.5 (over 100k) | Haiku 4.5 (previous gen) |
|---|---|---|---|
| Input | $0.10 | $0.50 | $1.00 |
| Output | $0.50 | $2.50 | $5.00 |
| Cache write (5-minute TTL) | $0.125 | $0.625 | $1.25 |
| Cache write (1-hour TTL) | $0.20 | $1.00 | $2.00 |
| Cache read (cache hit) | $0.01 | $0.05 | $0.10 |
| Batch API | 50% off input and output | 50% off | |
※ Haiku 5.5 figures are as published. The Haiku 4.5 cache figures are calculated by us from the official multipliers (1.25x for 5-minute writes, 2x for one hour, 0.1x for reads), verified 2026-10-09.
Source: Claude Haiku 5.5 — overview (Pricing / Availability), checked 2026-10-09.
Where the two-tier pricing actually bites
The rate switches the moment a single prompt crosses 100,000 tokens. Worked examples:
| Case | Haiku 5.5 | Claude Sonnet 5.5 |
|---|---|---|
| 200,000 input + 10,000 output tokens | $0.125 ($0.50/M × 0.2 + $2.50/M × 0.01) | $0.50 ($2/M × 0.2 + $10/M × 0.01) |
| 50,000 input + 5,000 output tokens | $0.0075 ($0.10/M × 0.05 + $0.50/M × 0.005) | $0.15 ($2/M × 0.05 + $10/M × 0.005) |
- Up to 100k the rate is 1/20 of Sonnet 5.5. For high-volume classification, routing and extraction, that is an order-of-magnitude difference.
- Above 100k it is still about a quarter of Sonnet 5.5. The honest framing is not “expensive above 100k” but “unusually cheap below it”.
- Reused long prefixes belong in the cache: $0.01/M read (against $0.10/M on Sonnet 5.5) — the pattern that fits RAG and agent system prompts.
Four things to watch when migrating from Haiku 4.5
1. About 30% more tokens
The same text counts as roughly 30% more tokens on the new tokenizer. A 10x lower unit price works out to about 7.7x cheaper in practice (1 ÷ (0.1 × 1.3)) — not 10x.
2. Thinking is on by default
Adaptive thinking, default effort
medium. Budget for extra output tokens, and steer depth with effort.3. No temperature parameters
Verbatim: “Omit temperature, top_p, and top_k, since a non-default value for any of them returns a 400 error.”
4. Thinking blocks are account-bound
Verbatim: “Its thinking blocks work only in the account that produced them, or in an account linked to it.” Multi-account setups cannot carry them across.
Specifications and model IDs
| Item | Value (official page) |
|---|---|
| Context window | 1M tokens |
| Max output | 128K tokens (300K on the Batch API with the output-300k-2026-03-24 beta header) |
| Thinking | Adaptive (on by default); default effort medium |
| Latency | Fastest (official comparison table) |
| Modalities | text, images → text |
| Knowledge cutoff | Jun 2026 (training data cutoff Jun 2026) |
| Status | Active (latest). Retirement not sooner than October 7, 2027. |
| Platform | Model ID |
|---|---|
| Claude API | claude-haiku-5-5 |
| Amazon Bedrock | anthropic.claude-haiku-5-5 |
| Google Cloud (Vertex AI) | claude-haiku-5-5 |
| Microsoft Foundry | claude-haiku-5-5 |
| Claude Platform on AWS | claude-haiku-5-5 |
What the docs say, verbatim
“Claude Haiku 5.5 is built for high-volume, latency-sensitive work such as classification, routing, extraction, and subagent tasks. It supports adaptive thinking with the effort parameter, a 1M token context window, and up to 128k output tokens. It uses the same newer tokenizer as Claude 4.7 and later models, so the same text counts as approximately 30% more tokens than on Claude Haiku 4.5.”
— Claude Platform Docs, “Claude Haiku 5.5 — overview” (checked 2026-10-09)
Pricing, specifications and model IDs above are as published. We do not fill gaps with estimates: where the vendor publishes no benchmark table, we say so.
Good fit, poor fit
Good: bulk classification, extraction, routing
Many short calls. Up to 100k tokens the input rate is $0.10/M against $2/M on Sonnet 5.5.
Good: subagents and preprocessing
The docs name subagent tasks explicitly — a natural front stage for a stronger model.
Good: latency-sensitive chat
Fastest latency in the official comparison table.
Poor: audit or contract decisions
Retention, operator access and training terms are not on the model page. Start from ZOA, ZDR and the four requirements.
Poor: pipelines that set temperature
Non-default
temperature, top_p or top_k returns a 400 error.Reconsider: 100k+ single-shot prompts
Rates become $0.50 / $2.50. For whole-document work, compare the total against Sonnet 5.5.
Related pages
- Claude Sonnet 5.5 — the balanced model of the same family ($2/$10, 1M context)
- Claude Opus 5.5 — the tier above ($4/$20)
- LLM API pricing comparison — every vendor on one page
- Token cost calculator — turn token counts into dollars
- Release timeline — where Haiku 5.5 sits
- Prompt caching explained — when $0.01/M cache reads pay off
⚠️ Disclaimer
- Prices and specifications were checked on 2026-10-09 against Claude Platform Docs (Haiku 5.5 overview and the official pricing documentation). Vendors change them without notice.
- Haiku 4.5 cache figures are our own calculation from the official multipliers.
- This site is informational and does not endorse or represent any provider.