AI inference fees-MaxCompute(MaxCompute)-阿里云帮助中心

The MaxCompute AI inference is a new, out-of-the-box, pay-as-you-go feature that lets you use large models for data processing or offline inference. This topic describes the billing rules for the AI inference.

Overview

The MaxCompute AI inference service provides large model inference, letting you call out-of-the-box models directly from SQL and MaxFrame jobs using a built-in AI function. You are billed on a pay-as-you-go basis according to the total number of input and output tokens.

Supported regions: The model inference service is available only in the following regions.
- Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), and China (Ulanqab)
- International: Singapore

Supported models: To use the following models, you must first enable the model inference service in the target region, such as China (Beijing).

Model	Type	Description
`qwen3.8-max`	Supports multimodal data input	A flagship Mixture of Experts (MoE) model from the Qwen 3.8 series with 2.4 trillion parameters. It features significantly enhanced programming and office productivity capabilities, and can autonomously program for over ten days to deliver complete projects. This model is ideal for highly complex scenarios such as autonomous programming, advanced office automation, scientific research, and long-duration tasks.
`qwen3.7-max`	Text input only	A flagship model from the Qwen 3.7 series, featuring comprehensive upgrades in inference, code generation, and multilingual understanding. It is suitable for highly complex tasks.
`qwen3.7-plus`	Supports multimodal data input	A balanced model from the Qwen 3.7 series that offers a good trade-off between performance and cost. It is suitable for enterprise scenarios such as long-text analysis and multi-turn conversations.
`qwen3.7-flash`	Supports multimodal data input	A lightweight, high-speed, native vision-language model from the Qwen 3.7 series. It features enhanced multimodal understanding and agent execution capabilities, making it ideal for high-concurrency, low-latency online services and lightweight multimodal tasks.
`qwen3.7-text-embedding`	Text input only	This unified multilingual text vector model, trained on Qwen 3.7, offers significant performance improvements in text retrieval, clustering, and classification compared to the text-embedding-v4 version.
`qwen3-vl-embedding`	Supports multimodal data input	A Qwen multimodal vector embedding model that supports vector representations for mixed image and text inputs. It is suitable for cross-modal retrieval and image-text matching scenarios.
`text-embedding-v4`	Text input only	A high-precision text vector embedding model suitable for semantic search, clustering, and similarity calculation.
`qwen3.6-plus`	Supports multimodal data input	A balanced model from the Qwen 3.6 series that offers a good trade-off between performance and cost. It is suitable for general business scenarios such as dialogue, summarization, and analysis.
`qwen3.6-flash`	Supports multimodal data input	A lightweight, high-speed model from the Qwen 3.6 series. It provides fast responses at a low cost, making it ideal for high-concurrency, low-latency online services.
`deepseek-v4-pro`	Text input only	The fourth-generation flagship inference model from DeepSeek. It excels at high-difficulty tasks such as mathematics, coding, and complex logical reasoning.
`deepseek-v4-flash`	Text input only	The fourth-generation lightweight model from DeepSeek. It offers fast inference and is highly cost-effective, making it suitable for daily conversations and lightweight inference tasks.
`qwen3.5-397b-a17b`	Supports multimodal data input	A Mixture of Experts (MoE) model from the Qwen 3.5 series. It features efficient parameter activation, a vast knowledge base, and multi-step reasoning capabilities. This model is designed for high-difficulty logical deduction, full-stack code generation, and in-depth knowledge Q&A.
`qwen3.5-omni-plus`	Supports multimodal data input	Qwen3.5-Omni is Qwen's latest omni-modal large model. It is suitable for scenarios such as text creation, voice assistants, and multimedia analysis, providing a natural and smooth multimodal understanding and interaction experience.
`qwen3-asr-flash`	Supports multimodal data input	A lightweight speech recognition model from the Qwen 3 series. It provides real-time transcription with low latency and high concurrency, making it suitable for real-time audio processing scenarios such as generating meeting minutes, live-streaming subtitles, and recognizing voice commands.
`tongyi-embedding-vision-plus`	Supports multimodal data input	Tongyi-Embedding-Vision is a visual multimodal representation model built on a large language model (LLM). It is suitable for a variety of downstream tasks, including image-to-image search, text-to-image search, text-to-video search, video-to-video search, text-to-text search, and text-to-image-and-text search.
`qwen3-max` (Sunsetting soon)	Text input only	A high-performance large language model from the Qwen series, suitable for complex inference and content generation. Important The `qwen3-max` model is sunsetting soon. To ensure service continuity, migrate your related services to a new model as soon as possible. For more information, see Model Decommissioning Policy.

Note

You incur model inference service fees only when you use the public models listed above in MaxFrame Overview and SQL jobs using an AI function. You are charged only for successful jobs; failed jobs do not incur fees.

Billing rules

The model inference service is billed based on token usage. The billing dimensions include:

Region
Model type
Token type: Distinguishes between input and output tokens and indicates the usage-based pricing tier.

Pricing

A single SQL or MaxFrame job may involve multiple model invocations. If the model has pricing tiers, the system calculates the input token count for each invocation separately to determine its billing tier (token type). The system then meters and bills usage separately for each token type.

Important

For deployments in China regions, compute resources for model inference are limited to the Chinese mainland, and static data is stored in the selected region.
For deployments in International regions, compute resources for model inference are dynamically scheduled worldwide (excluding the Chinese mainland), and static data is stored in the selected region.

qwen3.8-max

Chinese mainland

For deployments in China regions, compute resources for model inference are limited to the Chinese mainland, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Ulanqab)

Input tokens	Use case	Token type	Price (per million tokens)
0 < Tokens ≤ 1,048,576	model input	`input_token_tier1`	CNY 14.4
	model input (implicit cache hit)	`input_token_tier1_cached`	CNY 2.88
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	CNY 1.44
	model input (create explicit cache)	`input_token_tier1_create_cache`	CNY 18
	model output (non-thinking)	`output_token_tier1`	CNY 43.2
	model output (thinking mode)	`output_token_tier1_thinking`	CNY 43.2

International

For deployments in International regions, compute resources for model inference are dynamically scheduled worldwide (excluding the Chinese mainland), and static data is stored in the selected region.

Supported regions: Singapore

Input tokens	Use case	Token type	Price (per million tokens)
0 < Tokens ≤ 1,048,576	model input	`input_token_tier1`	CNY 17.9856
	model input (implicit cache hit)	`input_token_tier1_cached`	CNY 3.59712
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	CNY 1.79856
	model input (create explicit cache)	`input_token_tier1_create_cache`	CNY 22.482
	model output (non-thinking)	`output_token_tier1`	CNY 53.958
	model output (thinking mode)	`output_token_tier1_thinking`	CNY 53.958

Qwen3.7-max

Chinese mainland

For deployments in China regions, compute resources for model inference are limited to the Chinese mainland, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Ulanqab)

Input tokens	Use case	Token type	Price (per 1M tokens)
0 < tokens ≤ 1,048,576	Model input	`input_token_tier1`	CNY 14.4
	Model input (implicit cache hit)	`input_token_tier1_cached`	CNY 2.88
	Model input (explicit cache hit)	`input_token_tier1_cached_explicit`	CNY 1.44
	Model input (create explicit cache)	`input_token_tier1_create_cache`	CNY 18
	Model output	`output_token_tier1`	CNY 43.2
	Model output (thinking mode)	`output_token_tier1_thinking`	CNY 43.2

International

For deployments in International regions, compute resources for model inference are dynamically scheduled worldwide (excluding the Chinese mainland), and static data is stored in the selected region.

Supported region: Singapore

Input tokens	Use case	Token type	Price (per 1M tokens)
0 < tokens ≤ 1,048,576	Model input	`input_token_tier1`	CNY 22.4832
	Model input (implicit cache hit)	`input_token_tier1_cached`	CNY 4.49664
	Model input (explicit cache hit)	`input_token_tier1_cached_explicit`	CNY 2.24832
	Model input (create explicit cache)	`input_token_tier1_create_cache`	CNY 28.104
	Model output	`output_token_tier1`	CNY 67.4484
	Model output (thinking mode)	`output_token_tier1_thinking`	CNY 67.4484

qwen3.7-plus

Mainland China

For deployments in China regions, compute resources for model inference are limited to the Chinese mainland, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Ulanqab)

Tokens per inference	Scenario	Token type	Price / million tokens
0 < Token ≤ 262,144	model input	`input_token_tier1`	CNY 2.4
	model input (implicit cache hit) (implicit cache hit)	`input_token_tier1_cached`	CNY 0.48
	model input (explicit cache hit) (explicit cache hit)	`input_token_tier1_cached_explicit`	CNY 0.24
	model input (create explicit cache) (explicit cache creation)	`input_token_tier1_create_cache`	CNY 3
	model output (non-thinking)	`output_token_tier1`	CNY 9.6
	model output (thinking mode)	`output_token_tier1_thinking`	CNY 9.6
262,144 < Token ≤ 1,048,576	model input	`input_token_tier2`	CNY 7.2
	model input (implicit cache hit) (implicit cache hit)	`input_token_tier2_cached`	CNY 1.44
	model input (explicit cache hit) (Explicit cache hit)	`input_token_tier2_cached_explicit`	CNY 0.72
	model input (create explicit cache) (Explicit cache creation)	`input_token_tier2_create_cache`	CNY 9
	model output (non-thinking)	`output_token_tier2`	CNY 28.8
	model output (thinking mode)	`output_token_tier2_thinking`	CNY 28.8

International

For deployments in International regions, compute resources for model inference are dynamically scheduled worldwide (excluding the Chinese mainland), and static data is stored in the selected region.

Supported regions: Singapore

Tokens per inference	Scenario	Token type	Price / million tokens
0 < Token ≤ 262,144	model input	`input_token_tier1`	CNY 3.5976
	model input (implicit cache hit) (Implicit cache hit)	`input_token_tier1_cached`	CNY 0.71952
	model input (explicit cache hit) (explicit cache hit)	`input_token_tier1_cached_explicit`	CNY 0.35976
	model input (create explicit cache) (Explicit cache creation)	`input_token_tier1_create_cache`	CNY 4.497
	model output (non-thinking)	`output_token_tier1`	CNY 14.3892
	model output (thinking mode)	`output_token_tier1_thinking`	CNY 14.3892
262,144 < Token ≤ 1,048,576	model input	`input_token_tier2`	CNY 10.7916
	model input (implicit cache hit) (implicit cache hit)	`input_token_tier2_cached`	CNY 2.15832
	model input (explicit cache hit) (explicit cache hit)	`input_token_tier2_cached_explicit`	CNY 1.07916
	model input (create explicit cache) (Explicit cache creation)	`input_token_tier2_create_cache`	CNY 13.4895
	model output (non-thinking)	`output_token_tier2`	CNY 43.1664
	model output (thinking mode)	`output_token_tier2_thinking`	CNY 43.1664

qwen3.7-flash

Chinese mainland

For deployments in China regions, compute resources for model inference are limited to the Chinese mainland, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Ulanqab)

Input tokens	Usage scenario	Token type	Price (per million tokens)
0 < tokens ≤ 32,768	model input	`input_token_tier1`	CNY 0.24
	model input (implicit cache hit)	`input_token_tier1_cached`	CNY 0.048
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	CNY 0.024
	model input (create explicit cache)	`input_token_tier1_create_cache`	CNY 0.3
	model output (non-thinking)	`output_token_tier1`	CNY 0.96
	model output (thinking mode)	`output_token_tier1_thinking`	CNY 0.96
32,768 < tokens ≤ 262,144	model input	`input_token_tier2`	CNY 0.72
	model input (implicit cache hit)	`input_token_tier2_cached`	CNY 0.144
	model input (explicit cache hit)	`input_token_tier2_cached_explicit`	CNY 0.072
	model input (create explicit cache)	`input_token_tier2_create_cache`	CNY 0.9
	model output (non-thinking)	`output_token_tier2`	CNY 2.88
	model output (thinking mode)	`output_token_tier2_thinking`	CNY 2.88
262,144 < tokens ≤ 1,048,576	model input	`input_token_tier3`	CNY 1.44
	model input (implicit cache hit)	`input_token_tier3_cached`	CNY 0.288
	model input (explicit cache hit)	`input_token_tier3_cached_explicit`	CNY 0.144
	model input (create explicit cache)	`input_token_tier3_create_cache`	CNY 1.8
	model output (non-thinking)	`output_token_tier3`	CNY 5.76
	model output (thinking mode)	`output_token_tier3_thinking`	CNY 5.76

International

For deployments in International regions, compute resources for model inference are dynamically scheduled worldwide (excluding the Chinese mainland), and static data is stored in the selected region.

Supported regions: Singapore

Input tokens	Usage scenario	Token type	Price (per million tokens)
0 < tokens ≤ 32,768	model input	`input_token_tier1`	CNY 0.27
	model input (implicit cache hit)	`input_token_tier1_cached`	CNY 0.054
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	CNY 0.027
	model input (create explicit cache)	`input_token_tier1_create_cache`	CNY 0.3375
	model output (non-thinking)	`output_token_tier1`	CNY 1.1688
	model output (thinking mode)	`output_token_tier1_thinking`	CNY 1.1688
32,768 < tokens ≤ 262,144	model input	`input_token_tier2`	CNY 0.8988
	model input (implicit cache hit)	`input_token_tier2_cached`	CNY 0.17976
	model input (explicit cache hit)	`input_token_tier2_cached_explicit`	CNY 0.08988
	model input (create explicit cache)	`input_token_tier2_create_cache`	CNY 1.1235
	model output (non-thinking)	`output_token_tier2`	CNY 3.5976
	model output (thinking mode)	`output_token_tier2_thinking`	CNY 3.5976
262,144 < tokens ≤ 1,048,576	model input	`input_token_tier3`	CNY 1.7988
	model input (implicit cache hit)	`input_token_tier3_cached`	CNY 0.35976
	model input (explicit cache hit)	`input_token_tier3_cached_explicit`	CNY 0.17988
	model input (create explicit cache)	`input_token_tier3_create_cache`	CNY 2.2485
	model output (non-thinking)	`output_token_tier3`	CNY 7.194
	model output (thinking mode)	`output_token_tier3_thinking`	CNY 7.194

qwen3.7-text-embedding

The qwen3.7-text-embedding model is billed for input tokens only, with no charges for output tokens. This model has no pricing tiers, and all usage is billed as input_token_tier1.

For deployments in China regions, compute resources for model inference are limited to the Chinese mainland, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Ulanqab)

Tier	Use case	Token type	Unit price
No tiers	model input	`input_token_tier1`	CNY 0.6

qwen3-vl-embedding

For deployments in China regions, compute resources for model inference are limited to the Chinese mainland, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Ulanqab)

Region	Type	Use case	Token type	Unit price
Chinese mainland	text	model input	`input_token_tier1_text`	CNY 0.84
Chinese mainland	image and video	model input	`input_token_tier1_image`	CNY 2.16

text-embedding-v4

For text-embedding-v4, you are charged for model input tokens only. There are no charges for model output tokens and no pricing tiers. All usage is billed as input_token_tier1.

Chinese mainland

For deployments in China regions, compute resources for model inference are limited to the Chinese mainland, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Ulanqab)

Pricing tier	Use case	Token type	Price per million tokens
No tiers	Model input	`input_token_tier1`	CNY 0.6

International

For deployments in International regions, compute resources for model inference are dynamically scheduled worldwide (excluding the Chinese mainland), and static data is stored in the selected region.

Supported regions: Singapore

Pricing tier	Use case	Token type	Price per million tokens
No tiers	Model input	`input_token_tier1`	CNY 0.6168

qwen3.6-plus

Chinese mainland

For deployments in China regions, compute resources for model inference are limited to the Chinese mainland, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Ulanqab)

Input tokens	Scenario	Token type	Price (per million tokens)
0 < tokens ≤ 262,144	model input	`input_token_tier1`	CNY 2.4
	model input (implicit cache hit)	`input_token_tier1_cached`	CNY 0.48
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	CNY 0.24
	model input (create explicit cache)	`input_token_tier1_create_cache`	CNY 3
	model output (non-thinking)	`output_token_tier1`	CNY 14.4
	model output (thinking mode)	`output_token_tier1_thinking`	CNY 14.4
262,144 < tokens ≤ 1,048,576	model input	`input_token_tier2`	CNY 9.6
	model input (implicit cache hit)	`input_token_tier2_cached`	CNY 1.92
	model input (explicit cache hit)	`input_token_tier2_cached_explicit`	CNY 0.96
	model input (create explicit cache)	`input_token_tier2_create_cache`	CNY 12
	model output (non-thinking)	`output_token_tier2`	CNY 57.6
	model output (thinking mode)	`output_token_tier2_thinking`	CNY 57.6

International

For deployments in International regions, compute resources for model inference are dynamically scheduled worldwide (excluding the Chinese mainland), and static data is stored in the selected region.

Supported regions: Singapore

Input tokens	Scenario	Token type	Price (per million tokens)
0 < tokens ≤ 262,144	model input	`input_token_tier1`	CNY 4.49652
	model input (implicit cache hit)	`input_token_tier1_cached`	CNY 0.899304
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	CNY 0.449652
	model input (create explicit cache)	`input_token_tier1_create_cache`	CNY 5.62065
	model output (non-thinking)	`output_token_tier1`	CNY 26.97912
	model output (thinking mode)	`output_token_tier1_thinking`	CNY 26.97912
262,144 < tokens ≤ 1,048,576	model input	`input_token_tier2`	CNY 17.98608
	model input (implicit cache hit)	`input_token_tier2_cached`	CNY 3.597216
	model input (explicit cache hit)	`input_token_tier2_cached_explicit`	CNY 1.798608
	model input (create explicit cache)	`input_token_tier2_create_cache`	CNY 22.4826
	model output (non-thinking)	`output_token_tier2`	CNY 53.958
	model output (thinking mode)	`output_token_tier2_thinking`	CNY 53.958

qwen3.6-flash

Chinese mainland

For deployments in China regions, compute resources for model inference are limited to the Chinese mainland, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Ulanqab)

Input tokens	Use case	Token type	Price (per million tokens)
0 < tokens ≤ 262,144	model input	`input_token_tier1`	CNY 1.44
	model input (implicit cache hit)	`input_token_tier1_cached`	CNY 0.288
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	CNY 0.144
	model input (create explicit cache)	`input_token_tier1_create_cache`	CNY 1.8
	model output (non-thinking)	`output_token_tier1`	CNY 8.64
	model output (thinking mode)	`output_token_tier1_thinking`	CNY 8.64
262,144 < tokens ≤ 1,048,576	model input	`input_token_tier2`	CNY 5.76
	model input (implicit cache hit)	`input_token_tier2_cached`	CNY 1.152
	model input (explicit cache hit)	`input_token_tier2_cached_explicit`	CNY 0.576
	model input (create explicit cache)	`input_token_tier2_create_cache`	CNY 7.2
	model output (non-thinking)	`output_token_tier2`	CNY 34.56
	model output (thinking mode)	`output_token_tier2_thinking`	CNY 34.56

International

For deployments in International regions, compute resources for model inference are dynamically scheduled worldwide (excluding the Chinese mainland), and static data is stored in the selected region.

Supported regions: Singapore

Input tokens	Use case	Token type	Price (per million tokens)
0 < tokens ≤ 262,144	model input	`input_token_tier1`	CNY 2.24826
	model input (implicit cache hit)	`input_token_tier1_cached`	CNY 0.449652
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	CNY 0.224826
	model input (create explicit cache)	`input_token_tier1_create_cache`	CNY 2.810325
	model output (non-thinking)	`output_token_tier1`	CNY 13.48956
	model output (thinking mode)	`output_token_tier1_thinking`	CNY 13.48956
262,144 < tokens ≤ 1,048,576	model input	`input_token_tier2`	CNY 8.99304
	model input (implicit cache hit)	`input_token_tier2_cached`	CNY 1.798608
	model input (explicit cache hit)	`input_token_tier2_cached_explicit`	CNY 0.899304
	model input (create explicit cache)	`input_token_tier2_create_cache`	CNY 11.2413
	model output (non-thinking)	`output_token_tier2`	CNY 35.97096
	model output (thinking mode)	`output_token_tier2_thinking`	CNY 35.97096

deepseek-v4-pro

Chinese mainland

For deployments in China regions, compute resources for model inference are limited to the Chinese mainland, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Ulanqab)

Pricing tier	Use case	Token type	Unit price
No tiers	model input	`input_token_tier1`	CNY 14.4
	model input (implicit cache hit)	`input_token_tier1_cached`	CNY 2.88
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	CNY 1.44
	model input (create explicit cache)	`input_token_tier1_create_cache`	CNY 18
	model output (non-thinking)	`output_token_tier1`	CNY 28.8
	model output (thinking mode)	`output_token_tier1_thinking`	CNY 28.8

International

For deployments in International regions, compute resources for model inference are dynamically scheduled worldwide (excluding the Chinese mainland), and static data is stored in the selected region.

Supported regions: Singapore

Input tokens per inference	Use case	Token type	Unit price
0 < tokens ≤ 1,048,576	model input	`input_token_tier1`	CNY 21.5832
	model input (implicit cache hit)	`input_token_tier1_cached`	CNY 4.31664
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	CNY 2.15832
	model input (create explicit cache)	`input_token_tier1_create_cache`	CNY 26.979
	model output (non-thinking)	`output_token_tier1`	CNY 43.1664
	model output (thinking mode)	`output_token_tier1_thinking`	CNY 43.1664

deepseek-v4-flash

Chinese mainland

For deployments in China regions, compute resources for model inference are limited to the Chinese mainland, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Ulanqab)

Pricing tier	Use case	Token type	Unit price
No tiers	model input	`input_token_tier1`	CNY 1.2
	model input (implicit cache hit)	`input_token_tier1_cached`	CNY 0.24
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	CNY 0.12
	model input (explicit cache creation)	`input_token_tier1_create_cache`	CNY 1.5
	model output (non-thinking)	`output_token_tier1`	CNY 2.4
	model output (thinking mode)	`output_token_tier1_thinking`	CNY 2.4

International

For deployments in International regions, compute resources for model inference are dynamically scheduled worldwide (excluding the Chinese mainland), and static data is stored in the selected region.

Supported region: Singapore

Pricing tier	Use case	Token type	Unit price
0 < tokens ≤ 1,048,576	model input	`input_token_tier1`	CNY 1.7988
	model input (implicit cache hit)	`input_token_tier1_cached`	CNY 0.35976
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	CNY 0.17988
	model input (explicit cache creation)	`input_token_tier1_create_cache`	CNY 2.2485
	model output (non-thinking)	`output_token_tier1`	CNY 3.5976
	model output (thinking mode)	`output_token_tier1_thinking`	CNY 3.5976

qwen3.5-397b-a17b

For deployments in China regions, compute resources for model inference are limited to the Chinese mainland, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Ulanqab)

Tokens per inference	Usage scenario	Type	Unit price (per million tokens)
0 < tokens ≤ 131,072	model input	`input_token_tier1`	CNY 1.44
	model input (implicit cache hit)	`input_token_tier1_cached`	CNY 0.288
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	CNY 0.144
	model input (create explicit cache)	`input_token_tier1_create_cache`	CNY 1.8
	model output (non-thinking)	`output_token_tier1`	CNY 8.64
	model output (thinking mode)	`output_token_tier1_thinking`	CNY 8.64
131,072 < tokens ≤ 262,144	model input	`input_token_tier2`	CNY 3.6
	model input (implicit cache hit)	`input_token_tier2_cached`	CNY 0.72
	model input (explicit cache hit)	`input_token_tier2_cached_explicit`	CNY 0.36
	model input (create explicit cache)	`input_token_tier2_create_cache`	CNY 4.5
	model output (non-thinking)	`output_token_tier2`	CNY 21.6
	model output (thinking mode)	`output_token_tier2_thinking`	CNY 21.6

qwen3.5-omni-plus

Chinese mainland

For deployments in China regions, compute resources for model inference are limited to the Chinese mainland, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Ulanqab)

Price tier	Use case	Token type	Unit price
Text	model input	`input_token_tier1_text`	CNY 8.4
Image	model input	`input_token_tier1_image`	CNY 8.4
Video	model input	`input_token_tier1_video`	CNY 8.4
Audio	model input	`input_token_tier1_audio`	CNY 63.6
Text	model output	`output_token_tier1_text`	CNY 48
Audio	model output	`output_token_tier1_audio`	CNY 255.6

International

For deployments in International regions, compute resources for model inference are dynamically scheduled worldwide (excluding the Chinese mainland), and static data is stored in the selected region.

Supported region: Singapore

Price tier	Use case	Token type	Unit price
Text	model input	`input_token_tier1_text`	CNY 12.588
Image	model input	`input_token_tier1_image`	CNY 12.588
Video	model input	`input_token_tier1_video`	CNY 12.588
Audio	model input	`input_token_tier1_audio`	CNY 98.928
Text	model output	`output_token_tier1_text`	CNY 74.64
Audio	model output	`output_token_tier1_audio`	CNY 395.688

qwen3-asr-flash

Billing for qwen3-asr-flash is based only on model input tokens. There are no charges for model output and no pricing tiers. All usage is billed under input_token_tier1.

Chinese mainland

For deployments in China regions, compute resources for model inference are limited to the Chinese mainland, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Ulanqab)

Tier	Use case	Token type	Unit price
No tiers	model input	`input_token_tier1`	CNY 10.56

International

For deployments in International regions, compute resources for model inference are dynamically scheduled worldwide (excluding the Chinese mainland), and static data is stored in the selected region.

Supported regions: Singapore

Tier	Use case	Token type	Unit price
No tiers	model input	`input_token_tier1`	CNY 12.48

tongyi-embedding-vision-plus

For deployments in China regions, compute resources for model inference are limited to the Chinese mainland, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Ulanqab)

Pricing tier	Use case	Token type	Unit price (per million tokens)
Text	Model input	`input_token_tier1_text`	CNY 0.6
Image & video	Model input	`input_token_tier1_image`	CNY 0.6

qwen3-max (sunsetting soon)

Chinese mainland

For deployments in China regions, compute resources for model inference are limited to the Chinese mainland, and static data is stored in the selected region.

Supported regions:

Chinese mainland: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Ulanqab)

Tokens per inference	Scenario	Token type	Price (CNY per million tokens)
0 < tokens ≤ 32,768	model input	`input_token_tier1`	3
	model input (implicit cache hit)	`input_token_tier1_cached`	0.6
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	0.3
	model input (create explicit cache)	`input_token_tier1_create_cache`	3.75
	model output (standard)	`output_token_tier1`	12
	model output (thinking mode)	`output_token_tier1_thinking`	12
32,768 < tokens ≤ 131,072	model input	`input_token_tier2`	4.8
	model input (implicit cache hit)	`input_token_tier2_cached`	0.96
	model input (explicit cache hit)	`input_token_tier2_cached_explicit`	0.48
	model input (create explicit cache)	`input_token_tier2_create_cache`	6
	model output (standard)	`output_token_tier2`	19.2
	model output (thinking mode)	`output_token_tier2_thinking`	19.2
131,072 < tokens ≤ 258,048	model input	`input_token_tier3`	8.4
	model input (implicit cache hit)	`input_token_tier3_cached`	1.68
	model input (explicit cache hit)	`input_token_tier3_cached_explicit`	0.84
	model input (create explicit cache)	`input_token_tier3_create_cache`	10.5
	model output (standard)	`output_token_tier3`	33.6
	model output (thinking mode)	`output_token_tier3_thinking`	33.6

International

For deployments in International regions, compute resources for model inference are dynamically scheduled worldwide (excluding the Chinese mainland), and static data is stored in the selected region.

Supported region: Singapore

Tokens per inference	Scenario	Token type	Price (CNY per million tokens)
0 < tokens ≤ 32,768	model input	`input_token_tier1`	10.5684
	model input (implicit cache hit)	`input_token_tier1_cached`	2.11368
	model input (explicit cache hit)	`input_token_tier1_cached_explicit`	1.05684
	model input (create explicit cache)	`input_token_tier1_create_cache`	13.2105
	model output (standard)	`output_token_tier1`	52.842
	model output (thinking mode)	`output_token_tier1_thinking`	52.842
32,768 < tokens ≤ 131,072	model input	`input_token_tier2`	21.1368
	model input (implicit cache hit)	`input_token_tier2_cached`	4.22736
	model input (explicit cache hit)	`input_token_tier2_cached_explicit`	2.11368
	model input (create explicit cache)	`input_token_tier2_create_cache`	26.421
	model output (standard)	`output_token_tier2`	105.6852
	model output (thinking mode)	`output_token_tier2_thinking`	105.6852
131,072 < tokens ≤ 258,048	model input	`input_token_tier3`	26.4216
	model input (implicit cache hit)	`input_token_tier3_cached`	5.28432
	model input (explicit cache hit)	`input_token_tier3_cached_explicit`	2.64216
	model input (create explicit cache)	`input_token_tier3_create_cache`	33.027
	model output (standard)	`output_token_tier3`	132.1068
	model output (thinking mode)	`output_token_tier3_thinking`	132.1068

Billing

Billing frequency: Billed hourly.
Due to data aggregation delays, inference fees may take several hours to appear on your bill after a job is complete. The information in Alibaba Cloud Billing Management is final.
View your bills
1. Log on to the Alibaba Cloud Billing Management console.
2. In the left navigation pane, choose Billing > Bill Details.
3. On the Bill Details page, set Product Name to MaxCompute and Product Name to MaxCompute model computing service.

FAQ

Q: How do I estimate inference costs?
A: You can estimate costs based on the pricing tables and your expected input and output token lengths. For example, a job with 1,000 inference calls using the qwen3-max model, where each call averages 2,000 input and 1,000 output tokens, costs approximately CNY 18 for the model compute service.
Q: Is a free tier or trial available?
A: The model compute service is offered only on a pay-as-you-go basis and does not have a free tier. We recommend testing with a small dataset first to validate performance and estimate costs.
Q: Can I set a spending limit to prevent overages?
A: No, the service does not support spending or usage limits. To control costs, monitor your usage by reviewing your bills.

Overview

Billing rules

Pricing

qwen3.8-max

Chinese mainland

International

Qwen3.7-max

Chinese mainland

International

qwen3.7-plus

Mainland China

International

qwen3.7-flash

Chinese mainland

International

qwen3.7-text-embedding

qwen3-vl-embedding

text-embedding-v4

Chinese mainland

International

qwen3.6-plus

Chinese mainland

International

qwen3.6-flash

Chinese mainland

International

deepseek-v4-pro

Chinese mainland

International

deepseek-v4-flash

Chinese mainland

International

qwen3.5-397b-a17b

qwen3.5-omni-plus

Chinese mainland

International

qwen3-asr-flash

Chinese mainland

International

tongyi-embedding-vision-plus

qwen3-max (sunsetting soon)

Chinese mainland

International

Billing

FAQ

Related documentation