Pay-as-you-go pricing for model calls - EMR Serverless Spark - E-MapReduce

With pay-as-you-go billing, you are charged based on the token usage of built-in AI Center models in your workspace.

Note

AI Center became generally available on April 27, 2026 and is now billed. For details, see the EMR Serverless Spark AI Center general availability.

Billing details

Feature

Description

Billing rules

Fees are based on built-in model token usage per one-hour billing cycle. Rules vary by model:

qwen3.6-plus: Billed by input and output tokens. Same price for thinking and non-thinking modes.
qwen3.5-plus: Billed by input and output tokens. Same price for thinking and non-thinking modes.
qwen-plus: Billed by input and output tokens. Prices differ between thinking and non-thinking modes.
text-embedding-v4: Billed by input tokens only.
tongyi-embedding-vision-plus: Billed by input tokens only. Supports image, video, and text inputs.

Model call cost formula: Input token usage × Input unit price + Output token usage × Output unit price

For example, if you make 10,000 ai_query() calls in Singapore, and each call has 260 input tokens and 50 output tokens (in non-thinking mode), the total cost is: 0.48 × 260 × 10000 ÷ 1000000 + 1.44 × 50 × 10000 ÷ 1000000 = 0.8448 USD.

Note

To estimate token usage, see model call.

Billing cycle

Fees are calculated hourly (UTC+8). After each cycle, the system generates a bill and deducts fees from your account. Bill data may be delayed. Understand your bill.

Regional unit prices

qwen3.6-plus

Note

qwen3.6-plus uses the same prices for thinking and non-thinking modes.

Region	Input token range	Input unit price (USD/million tokens)	Output unit price (USD/million tokens)
China (Beijing) China (Shanghai) China (Hangzhou) China (Shenzhen)	0 < Token ≤ 128K	0.331	1.981
	128K < Token ≤ 256K	1.321	7.927
China (Hong Kong) Singapore Germany (Frankfurt) US (Virginia) US (Silicon Valley) Japan (Tokyo) Indonesia (Jakarta) Mexico	0 < Token ≤ 256K	0.6	3.6
	256K < Token ≤ 1M	2.4	7.2

qwen3.5-plus

Note

qwen3.5-plus uses the same prices for thinking and non-thinking modes.

Region	Input token range	Input unit price (USD/million tokens)	Output unit price (USD/million tokens)
China (Beijing) China (Shanghai) China (Hangzhou) China (Shenzhen)	0 < Token ≤ 128K	0.138	0.826
	128K < Token ≤ 256K	0.344	2.064
	256K < Token ≤ 1M	0.688	4.128
China (Hong Kong) Singapore Germany (Frankfurt) US (Virginia) US (Silicon Valley) Japan (Tokyo) Indonesia (Jakarta) Mexico	0 < Token ≤ 256K	0.48	2.88
	256K < Token ≤ 1M	0.6	3.6

qwen-plus

Region	Mode	Input token range per request	Input unit price (USD/million tokens)	Output unit price (USD/million tokens)
China (Beijing) China (Shanghai) China (Hangzhou) China (Shenzhen)	non-thinking mode	0 < Token ≤ 128K	0.138	0.344
		128K < Token ≤ 256K	0.414	3.442
		256K < Token ≤ 1M	0.827	8.257
	thinking mode	0 < Token ≤ 128K	0.138	1.376
		128K < Token ≤ 256K	0.414	4.130
		256K < Token ≤ 1M	0.827	11.009
China (Hong Kong) Singapore Germany (Frankfurt) US (Virginia) US (Silicon Valley) Japan (Tokyo) Indonesia (Jakarta) Mexico	non-thinking mode	0 < Token ≤ 256K	0.48	1.44
	non-thinking mode	256K < Token ≤ 1M	1.44	4.32
	thinking mode	0 < Token ≤ 256K	0.48	4.80
	thinking mode	256K < Token ≤ 1M	1.44	14.40

text-embedding-v4

Region	Input unit price (USD/million tokens)
China (Beijing) China (Shanghai) China (Hangzhou) China (Shenzhen)	0.086
China (Hong Kong) Singapore Germany (Frankfurt) US (Virginia) US (Silicon Valley) Japan (Tokyo) Indonesia (Jakarta) Mexico	0.084

tongyi-embedding-vision-plus

Region	Input modality	Input unit price (USD/million tokens)
Indonesia (Jakarta)	image/video/text	0.09