Haiku 4.5 or 5.5? $0.10/$0.50 Specs and NT$ Cost Estimates

Data checked on October 8, 2026. Specs and prices come from the BazaarLink public model catalog (/api/v1/models), and benchmark scores come from Artificial Analysis (AA). Both were checked on the same day.

Exchange rate: the NT$ examples use the public rate of October 8, 2026, US$1 = NT$31.8690 (ExchangeRate-API, data updated 2026-10-08 00:02 UTC). NT$ figures are rounded and for estimates only; they are not the amount a credit card is actually charged.

Claude Haiku 5.5 (model ID claude-haiku-5.5) is now available on BazaarLink.

The short version

• The unit price is one tenth of Haiku 4.5. Haiku 5.5 costs $0.10 per million input tokens and $0.50 output; Haiku 4.5 costs $1 / $5. For the same usage the bill is 90% lower.
• Context grows from 200,000 to 1,000,000 tokens, with one price threshold. When a single request's input exceeds 100,000 tokens, that whole request is charged $0.50 input and $2.50 output (5x the base price). Even at that tier, the price is still half of Haiku 4.5's.
• The AA Intelligence Index is clearly higher. Haiku 5.5 scores 34 at the default medium reasoning effort and 43 at the highest, Max; AA lists Claude 4.5 Haiku (Reasoning) at 17. A score is a model × reasoning-effort combination, so scores at different efforts cannot be compared directly, and they do not mean your task will improve by the same amount.
• Because it is cheaper, the cost of switching is testing time, not the bill. Haiku 4.5 is still in the model catalog, so pick a set of representative tasks, test side by side, and then decide whether to switch everywhere.

Specs and prices

The surcharge above 100,000 tokens swaps the whole price list at once (input and output both become 5x), and it is decided by the number of input tokens in a single request. A request of exactly 100,000 tokens is still billed at the base price. The BazaarLink catalog describes Haiku 5.5 as a small, fast model for high-volume, cost-sensitive work such as summarization, subagents and browser use. For live prices, see the BazaarLink model catalog.

NT$ cost examples

Per-million-token public prices converted at the rate above:

A monthly customer-support bot example: 500 conversations a day, 3 turns each, 300 input + 150 output tokens per turn, which is about 13.5 million input and 6.75 million output tokens a month. Haiku 4.5 costs 13.5 × US$1 + 6.75 × US$5 = US$47.25, about NT$1,506; Haiku 5.5 costs 13.5 × US$0.10 + 6.75 × US$0.50 = US$4.725, about NT$151. Every turn's input is far below 100,000 tokens, so the surcharge threshold is never reached. Your real monthly cost depends on your own usage logs.

What Artificial Analysis says

AA is an independent model benchmarking site. As of October 8, 2026, AA's Intelligence Index for Haiku 5.5 is below (each reasoning effort is scored separately), with AA's Haiku 4.5 for reference:

AA lists Claude 4.5 Haiku (Reasoning) at an Intelligence Index of 17. Keep in mind:

• A score is a model × reasoning-effort combination, so different efforts cannot be compared directly. Haiku 4.5 and 5.5 reasoning settings also do not necessarily map one to one, so this comparison only shows the general direction.
• The Intelligence Index is AA's own benchmark mix. It does not mean your tasks, language or tool flow will improve by the same amount.
• All numbers in the table come from AA's Haiku 5.5 pages and may change later. AA's pages show whole numbers.

Who should switch?

Try Haiku 5.5 first if:

• You use Haiku 4.5 for high-volume, low-complexity work (classification, summarization, data extraction, customer-support FAQ) and cost is the main pressure.
• You need more than 200,000 tokens of context and want a low unit price: Haiku 4.5 cannot do a 1,000,000-token context.
• It is a new project: the catalog still has Haiku 4.5, but there is no price reason to pick the old version in this price band.

Test before switching if:

• Your flow depends on the existing output format or tone, and you need to confirm the new version does not break downstream parsing.
• You often send inputs over 100,000 tokens: a single request is then billed at 5x, so estimate with your real input length first and confirm it is still worth it.

Suggested switching process: pick 20 to 30 real tasks, fix the prompts, tools and reasoning effort, run Haiku 4.5 and 5.5 together, and record pass rate, human correction time and input/output tokens. Switch everywhere once you are satisfied.

If you are comparing other models, see Claude API NT$ cost estimates (Traditional Chinese), Sonnet 5 to Sonnet 5.5, and Haiku 5.5 vs DeepSeek, GLM, MiMo Flash: price and AA index.

FAQ

How much cheaper is Haiku 5.5 than Haiku 4.5?

Input drops from $1 to $0.10 per million tokens and output from $5 to $0.50, both one tenth. Even when input goes over 100,000 tokens and the surcharge applies (input $0.50, output $2.50), it is still half of Haiku 4.5's price.

How long is Haiku 5.5's context? Does the price rise with length?

The context length is 1,000,000 tokens. When a single request's input exceeds 100,000 tokens, that whole request is charged $0.50 input and $2.50 output; at or below that it is charged at the base price of $0.10 / $0.50.

Can I keep using Haiku 4.5?

Yes. Haiku 4.5 is still in the BazaarLink model catalog. The new version is cheaper, so you can test side by side first and do not need to switch everything at once.

Does Haiku 5.5 support image and file input?

Yes. Input can be text, images and files; output is text.

Can I adjust Haiku 5.5's reasoning effort? Can I turn it off?

The public catalog lists five efforts: max, xhigh, high, medium and low. The default is medium, and reasoning is marked as not mandatory (mandatory: false). For the exact parameter syntax, follow the API documentation.

How often are the scores and prices in this article updated?

Prices and specs follow the BazaarLink public model catalog and AA scores follow the AA site. This article was checked on October 8, 2026; for later changes, refer to the live pages.

TWD billing · Taiwan invoices · leading AI models · OpenAI-compatible API