Agentic Search billing
Effective 00:00:00 on August 31, 2026 (Beijing time), the Agentic Search search agent becomes commercially available and is no longer free of charge as it was during the public beta. This topic describes the billing methods, billable items, and prices of Agentic Search.
Billing methods
All features of OpenSearch Agentic Search are settled on a pay-as-you-go basis. You are charged based on the number of tokens consumed or the number of calls made by the Agentic Search main pipeline and its dependent services. You are not charged if you do not use the service. Consistent with other services of the AI Search Open Platform, a bill is generated on an hourly basis. All hourly bills belong to the same order, and the fees are deducted from your Alibaba Cloud account based on the bill at the time the order is generated.
A pay-as-you-go fee consists of the following parts:
Pay-as-you-go fee = Number of input tokens x Input price + Number of output tokens x Output price + Number of web search calls x Web search priceDuring the public beta before 00:00:00 on August 31, 2026 (Beijing time), Agentic Search is free of charge. You are not charged and no token quota is consumed, except for specific service items.
Billable items
Agentic Search uses a combined billing model. In addition to the token consumption of the Agentic Search main pipeline, the services that are actually called during task execution, such as knowledge base parsing, document parsing and splitting, multimodal parsing, vectorization, web search, and Agentic Memory, are billed separately based on their own billable items and are itemized in your bill.
The following table describes the billable items and prices of each Agentic Search feature:
|
Feature |
Billable item |
Price (CNY) |
|
Agentic Search main pipeline |
agentic-search/output_token |
0.012 CNY/1,000 tokens |
|
agentic-search/input_token |
0.003 CNY/1,000 tokens |
|
|
text-generation/deepseek-v4-pro/input_token |
0.0144 CNY/1,000 tokens |
|
|
text-generation/deepseek-v4-pro/output_token |
0.0288 CNY/1,000 tokens |
|
|
text-generation/deepseek-v4-flash/input_token |
0.0012 CNY/1,000 tokens |
|
|
text-generation/deepseek-v4-flash/output_token |
0.0024 CNY/1,000 tokens |
|
|
text-generation/qwen3.7-plus/input_token |
0.0024 CNY/1,000 tokens |
|
|
text-generation/qwen3.7-plus/output_token |
0.0096 CNY/1,000 tokens |
|
|
text-generation/qwen3.6-flash/input_token |
0.00144 CNY/1,000 tokens |
|
|
text-generation/qwen3.6-flash/output_token |
0.00864 CNY/1,000 tokens |
|
|
text-generation/qwen3-vl-flash/input_token |
0.00018 CNY/1,000 tokens |
|
|
text-generation/qwen3-vl-flash/output_token |
0.0018 CNY/1,000 tokens |
|
|
text-generation/qwen3.5-plus/input_token |
0.00096 CNY/1,000 tokens |
|
|
text-generation/qwen3.5-plus/output_token |
0.00576 CNY/1,000 tokens |
|
|
text-generation/qwen3.6-35b-a3b/input_token |
0.00216 CNY/1,000 tokens |
|
|
text-generation/qwen3.6-35b-a3b/output_token |
0.01296 CNY/1,000 tokens |
|
|
text-generation/qwen3.6-27b/input_token |
0.0036 CNY/1,000 tokens |
|
|
text-generation/qwen3.6-27b/output_token |
0.0216 CNY/1,000 tokens |
|
|
text-generation/kimi-k2.6/input_token |
0.0078 CNY/1,000 tokens |
|
|
text-generation/kimi-k2.6/output_token |
0.0324 CNY/1,000 tokens |
|
|
text-generation/kimi-k2.5/input_token |
0.0048 CNY/1,000 tokens |
|
|
text-generation/kimi-k2.5/output_token |
0.0252 CNY/1,000 tokens |
|
|
Knowledge base parsing and query |
agentic-search/knowledge-base/input_token |
0.0024 CNY/1,000 tokens |
|
agentic-search/knowledge-base/output_token |
0.012 CNY/1,000 tokens |
|
|
ranker/qwen3-rerank/token |
0.0006 CNY/1,000 tokens |
|
|
video-summarization/ops-video-summarize-embodied-001/output_token |
0.00768 CNY/1,000 tokens |
|
|
video-summarization/ops-video-summarize-embodied-001/input_token |
0.00192 CNY/1,000 tokens |
|
|
text-embedding/text-embedding-v4/token |
0.00062 CNY/1,000 tokens |
|
|
text-embedding/text-embedding-v3/token |
0.00062 CNY/1,000 tokens |
|
|
video-segmentation/ops-video-segment-001/image_token |
0.00216 CNY/1,000 tokens |
|
|
text-generation/qwen3-vl-plus/input_token |
0.0018 CNY/1,000 tokens |
|
|
text-generation/qwen3-vl-plus/output_token |
0.018 CNY/1,000 tokens |
|
|
Web search |
web-search/ops-web-search-001/lite/pv |
12 CNY/1,000 calls |
|
web-search/ops-web-search-001/pro/pv |
36 CNY/1,000 calls |
|
|
web-search/ops-web-search-001/pro-fetch/pv |
46 CNY/1,000 calls |
The token consumption generated when services such as Agentic Memory call large language models is settled based on the billable items of the corresponding text-generation models in the preceding table.
Precautions
-
The underlying retrieval engines on which Agentic Search depends (Alibaba Cloud Elasticsearch and OpenSearch Vector Search Edition) are billed separately. Their fees are not included in the preceding table. For more information, see the billing topics of the corresponding services.
-
This topic describes only the billable items involved in Agentic Search. For the billable items of other models and services of the AI Search Open Platform, the tiered pricing rules, and the prices on the international site, see Billing methods and billable items.
-
If your account has overdue payments, calls to Agentic Search are affected. Pay attention to your account balance and consumption in a timely manner.