How to check model training and deployment pricing-Alibaba Cloud Model Studio(Model Studio)-阿里云帮助中心

This topic describes the billing rules and pricing for model training and model deployment on Alibaba Cloud Model Studio.

Training billing

Text generation models – Qwen

Note

For the training workflow, see Model tuning. After training completes, deploy the new model before evaluating or calling it.

Method

Billed by training tokens

Formula

Model training fee = (Total tokens in training data + Total tokens in mixed training data) × Number of epochs × Training unit price (Minimum billing unit: 1 token)

View the estimated training fee at the bottom of the model training console, and click Computing Details to view the total number of training tokens, number of epochs, and training unit price.

Qwen

Model service	Model identifier	Price
Qwen3.6-Flash-2026-04-16	qwen3.6-flash-2026-04-16	CNY 0.05 per 1,000 tokens

Qwen3.5-27B	qwen3.5-27b	CNY 0.05 per 1,000 tokens
Qwen3.5-9B	qwen3.5-9b	CNY 0.02 per 1,000 tokens
Qwen3.5-Flash-2026-02-23	qwen3.5-flash-2026-02-23	CNY 0.05 per 1,000 tokens

Qwen3-32B	qwen3-32b	CNY 0.04 per 1,000 tokens
Qwen3-30B-A3B-Instruct-2507	qwen3-30b-a3b-instruct-2507	CNY 0.03 per 1,000 tokens
Qwen3-14B	qwen3-14b	CNY 0.03 per 1,000 tokens
Qwen3-8B	qwen3-8b	CNY 0.006 per 1,000 tokens
Qwen3-1.7B	qwen3-1.7b	CNY 0.0045 per 1,000 tokens
Qwen3-0.6B	qwen3-0.6b	CNY 0.003 per 1,000 tokens

Qwen2.5-72B-Instruct	qwen2.5-72b-instruct	CNY 0.15 per 1,000 tokens
Qwen2.5-32B-Instruct	qwen2.5-32b-instruct	CNY 0.03 per 1,000 tokens
Qwen2.5-14B-Instruct	qwen2.5-14b-instruct	CNY 0.03 per 1,000 tokens
Qwen2.5-7B-Instruct	qwen2.5-7b-instruct	CNY 0.006 per 1,000 tokens

Qwen-Plus-Character-2025-11-06	qwen-plus-character-2025-11-06	CNY 0.15 per 1,000 tokens

Qwen-VL

Model service	Model identifier	Price
Qwen3-VL-8B-Instruct	qwen3-vl-8b-instruct	CNY 0.012 per 1,000 tokens
Qwen3-VL-8B-Thinking	qwen3-vl-8b-thinking	CNY 0.012 per 1,000 tokens
Qwen3-VL-4B-Instruct	qwen3-vl-4b-instruct	CNY 0.006 per 1,000 tokens

Qwen2.5-VL-72B-Instruct	qwen2.5-vl-72b-instruct	CNY 0.05 per 1,000 tokens
Qwen2.5-VL-32B-Instruct	qwen2.5-vl-32b-instruct	CNY 0.02 per 1,000 tokens
Qwen2.5-VL-7B-Instruct	qwen2.5-vl-7b-instruct	CNY 0.01 per 1,000 tokens

Image generation models – Wan

Note

For the training workflow, see Fine-tune image generation models. After training completes, deploy the new model before calling it.

Method	Billed by training tokens
Formula	Model training fee = Total training tokens × Training unit price (Billing unit: per 1,000 tokens)

Formula for total training tokens

$训练 T o k e n 总量 \approx ma x_s t e p s \times L_{s t e p}$

Where:

max_steps: A hyperparameter specified during training, representing the maximum number of training steps (configured when creating a fine-tuning job).
L_step: The token consumption per step. The formula is:
$L_{s t e p} = i \in ba t c h \sum L_{i t e m}^{(i)} \leq L_{ma x}$
L_step is approximately equal to L_max. L_max is determined by the max_token_length and generation_type, as shown below:

generation_type	max_token_length	L_max
t2i (text-to-image)	1k	128,000
t2i (text-to-image)	2k	232,200
i2i (image-to-image)	1k	232,200
i2i (image-to-image)	2k	320,000

Note

The above formula provides an approximation. Actual billing is based on the usage field returned by the system.

Model	Code	Training price (per 1K tokens)
Wan image generation	wan2.7-image-pro	CNY 0.08
Wan image generation	wan2.7-image	CNY 0.08

Billing example

Suppose you fine-tune the wan2.7-image-pro model for t2i. The parameters are: max_steps = 200, max_token_length = "1k", and the training price is CNY 0.08 per 1,000 tokens:

From the table: L_max = 128,000 (generation_type=t2i, max_token_length=1k), L_step ≈ L_max = 128,000
Total training tokens ≈ 200 × 128,000 = 25,600,000 = 25,600 thousand tokens
Model training fee ≈ 25,600 × 0.08 = CNY 2,048

Video generation models – Wan

Note

For the training workflow, see Fine-tuning video generation models. After training completes, deploy the new model before calling it.

Method	Billed by training tokens
Formula	Model training fee = Total training tokens × Training unit price (Billing unit: per 1,000 tokens)

Formula for total training tokens

$T r ainin g T o k e n s T o t a l = (i = 1 \sum N billing duration of video_{i}) \times \frac{ma x _ p i x e l s}{1024} \times n_e p oc h s$

Where:

N: Total number of videos in the training set.
max_pixels: A hyperparameter specified during training, representing the maximum number of pixels for a video (configured when creating a fine-tuning job).
n_epochs: A hyperparameter specified during training, representing the number of loops (configured when creating a fine-tuning job).
Billing duration calculation rule for a single video: First, round the original video duration (in seconds) to the nearest integer, then determine the final value based on model limits.
- wan2.5 model: Billing duration=min(10, rounded duration), meaning a single video is billed for a maximum of 10 seconds.
- wan2.2 model: Billing duration=min(5, rounded duration), meaning a single video is billed for a maximum of 5 seconds.

Model	Code	Training price (per 1K tokens)
Wan image-to-video (first frame-based)	wan2.2-i2v-flash	CNY 0.06
Wan image-to-video (first frame-based)	wan2.5-i2v-preview	CNY 0.32
Image-to-video (first and last frame-based)	wan2.2-kf2v-flash	CNY 0.06

Billing example

Assume a training set contains two videos with durations of 3.4 seconds and 6.5 seconds. The parameters are set as follows: max_pixels = 262,144, n_epochs = 400, and the training price is CNY 0.06 per 1,000 tokens:

Duration calculation:
- Video 1: 3.4 seconds is rounded to 3 seconds. The billable duration is min(5, 3) = 3.
- Video 2: 6.5 seconds is rounded to 7 seconds. The billable duration is min(5, 7) = 5.
- Total billable duration = 3 + 5 = 8 seconds.
The total training tokens are calculated as 8 × (262144/1024) × 400, resulting in 819,200, or 819.2 thousand tokens.
Model training fee = 819.2 × 0.06 = CNY 49.152.

Deployment billing

Text generation models: Qwen

Pay-as-you-go

cost = usage duration × (input TPM price × input TPM + output TPM price × output TPM)

For pay-as-you-go billing, usage is calculated by the hour, and the "1-hour duration" column lists the unit price. For subscription billing, usage is calculated by the day, and the "1-day duration" column lists the unit price.

A subscription activates upon payment and is valid for N days, expiring at 23:59 on the Nth day. For orders placed after 22:00, the expiration date is automatically extended by 1 day.
After a subscription expires, service is suspended following a 2-hour grace period. Resources are then retained for 14 hours before being released.
You cannot terminate a subscription early.
For pay-as-you-go, if an account has an overdue balance, its deployed resources are retained and billing continues for 24 hours before they are automatically released.

When a model's input exceeds its maximum token limit or purchased TPM, calls to that model automatically switch to the pay-as-you-go mode. In this mode, inference performance may decrease, rate limiting is subject to the public traffic limits of the current snapshot model in your workspace, and fees are charged at the standard pay-as-you-go rate.

The API response headers will include x-dashscope-ptu-overflow:true.
To view TPM statistics, go to the Model Studio console.

See Configuration Downgrade Refund Rules for details on fee adjustments and refunds for a scale-in (configuration downgrade).

Qwen

Model name	Model code	Maximum input tokens	Pay-as-you-go input (per 10K TPM/hour)	Pay-as-you-go output (per 1K TPM/hour)	Provisioned input (per 10K TPM/day)	Provisioned output (per 1K TPM/day)
Qwen3.7-Max-2026-05-20	qwen3.7-max-2026-05-20	128,000	¥28.8	¥8.64	¥345.6	¥103.68

Qwen3.6-Flash-2026-04-16	qwen3.6-flash-2026-04-16	128,000	¥2.88	¥1.73	¥34.56	¥20.74
Qwen3.6-Plus-2026-04-02	qwen3.6-plus-2026-04-02	128,000	¥4.8	¥2.88	¥57.6	¥34.56

Qwen3.5-Plus-2026-04-20	qwen3.5-plus-2026-04-20	128,000	¥1.92	¥1.15	¥23.04	¥13.82

Qwen3-Max-2025-09-23	qwen3-max-2025-09-23	128,000	¥7.68	¥3.08	¥92.16	¥36.96

Qwen-Flash-2025-07-28	qwen-flash-2025-07-28	128,000	¥0.36	¥0.36	¥4.32	¥4.32
Qwen-Plus-2025-12-01	qwen-plus-2025-12-01	128,000	¥1.92	standard: ¥0.48 code interpreter: ¥1.92	¥23.04	standard: ¥5.76 code interpreter: ¥23.04

DeepSeek

Model name	Model code	Maximum input tokens	Pay-as-you-go input (per 10K TPM/hour)	Pay-as-you-go output (per 1K TPM/hour)	Provisioned input (per 10K TPM/day)	Provisioned output (per 1K TPM/day)
DeepSeek-v4-Pro	deepseek-v4-pro	64,000	¥43.2	¥8.64	¥518.4	¥103.68
DeepSeek-v3.2	deepseek-v3.2	64,000	¥7.2	¥1.08	¥86.4	¥12.96
DeepSeek-v3	deepseek-v3	64,000	¥7.2	¥2.88	¥86.4	¥34.56

Qwen-VL

Model name

Model code

Maximum input tokens

Pay-as-you-go input

(per 10K TPM/hour)

Pay-as-you-go output

(per 1K TPM/hour)

Provisioned input

(per 10K TPM/day)

Provisioned output

(per 1K TPM/day)

Qwen3-VL-Plus-2025-09-23

qwen3-vl-plus-2025-09-23

128,000

¥2.4

¥28.8

Other models

Model name

Model code

Maximum input tokens

Pay-as-you-go input

(per 10K TPM/hour)

Pay-as-you-go output

(per 1K TPM/hour)

Provisioned input

(per 10K TPM/day)

Provisioned output

(per 1K TPM/day)

GLM-5.1

glm-5.1

64,000

¥21.6

¥8.64

¥259.2

¥103.68

Billing by duration

fee = usage (hours) × number of model units × model unit price

For pay-as-you-go, the model unit price is the same as the hourly unit price in the table below. For monthly subscriptions, the formula is number of subscription months × number of model units × monthly unit price.

If you cancel a prepaid purchase within the first month, you will be billed at 1.2 times the daily rate (approximately the monthly rate / 30). Each partial day is billed as a full day.

Note

We allocate computing resources for pay-as-you-go model units on a first-come, first-served basis. If a purchase fails, we issue a full refund.

Text generation

Qwen

Model name	Model code	Unit	Hourly price (CNY)	Monthly price (CNY)
Qwen3.6-35B-A3B	qwen3.6-35b-a3b	MU8 x 1	¥47	¥22,400
Qwen3.6-35B-A3B	qwen3.6-35b-a3b	MU9 x 1	¥51	¥24,600
Qwen3.6-27B	qwen3.6-27b	MU9 x 1	¥51	¥24,600
Qwen3.6-Flash-2026-04-16	qwen3.6-flash-2026-04-16	MU1 x 2	¥108	¥52,236
Qwen3.6-Plus-2026-04-02	qwen3.6-plus-2026-04-02	MU1 x 8 MU1 x 16 (PD-separated mode)	¥432 ¥864	¥208,944 ¥417,888

Qwen3.5-397B-A17B	qwen3.5-397b-a17b	MU2 x 8	¥504	¥240,288
		MU3 x 8 MU3 x 16 (PD-separated mode)	¥1,096 ¥2,192	¥527,752 ¥1,055,504
		MU6 x 16	¥400	¥193,424
Qwen3.5-122B-A10B	qwen3.5-122b-a10b	MU1 x 4	¥216	¥104,472
		MU2 x 8	¥504	¥240,288
		MU6 x 16	¥400	¥193,424
		MU9 x 2	¥102	¥49,200
Qwen3.5-35B-A3B	qwen3.5-35b-a3b	MU1 x 2	¥108	¥52,236
		MU2 x 8	¥504	¥240,288
		MU8 x 1	¥47	¥22,400
		MU9 x 1	¥51	¥24,600
Qwen3.5-27B	qwen3.5-27b	MU1 x 2	¥108	¥52,236
Qwen3.5-27B	qwen3.5-27b	MU9 x 1	¥51	¥24,600
Qwen3.5-9B	qwen3.5-9b	MU1 x 2	¥108	¥52,236
		MU8 x 1	¥47	¥22,400
		MU9 x 1	¥51	¥24,600
Qwen3.5-Flash-2026-02-23	qwen3.5-flash-2026-02-23	MU1 x 2	¥108	¥52,236
Qwen3.5-Plus-2026-02-15	qwen3.5-plus-2026-02-15	MU1 x 16 (PD-separated mode)	¥864	¥417,888
Qwen3.5-Plus-2026-02-15	qwen3.5-plus-2026-02-15	MU3 x 8 MU3 x 16 (PD-separated mode)	¥1,096 ¥2,192	¥527,752 ¥1,055,504

Qwen3-235B-A22B-Instruct-2507	qwen3-235b-a22b-instruct-2507	MU1 x 4	¥216	¥104,472
Qwen3-235B-A22B-Instruct-2507	qwen3-235b-a22b-instruct-2507	MU2 x 8	¥504	¥240,288
Qwen3-Next-80B-A3B-Instruct	qwen3-next-80b-a3b-instruct	MU1 x 2	¥108	¥52,236
Qwen3-32B	qwen3-32b	MU1 x 4	¥216	¥104,472
Qwen3-32B	qwen3-32b	MU6 x 4	¥100	¥48,356
Qwen3-30B-A3B	qwen3-30b-a3b	MU9 x 2	¥102	¥49,200
Qwen3-30B-A3B-Instruct-2507	qwen3-30b-a3b-instruct-2507	MU1 x 4	¥216	¥104,472
Qwen3-30B-A3B-Instruct-2507	qwen3-30b-a3b-instruct-2507	MU2 x 8	¥504	¥240,288
Qwen3-8B	qwen3-8b	MU1 x 2	¥108	¥52,236
		MU2 x 2	¥126	¥60,072
		MU5 x 1	¥21	¥10,139
Qwen3-4B	qwen3-4b	MU1 x 2	¥108	¥52,236
Qwen3-4B	qwen3-4b	MU5 x 1	¥21	¥10,139
Qwen3-1.7B	qwen3-1.7b	MU1 x 2	¥108	¥52,236
Qwen3-1.7B	qwen3-1.7b	MU5 x 1	¥21	¥10,139
Qwen3-Embedding-0.6B	qwen3-embedding-0.6b	MU5 x 1	¥21	¥10,139
Qwen3-Embedding-0.6B	qwen3-embedding-0.6b	MU6 x 1	¥25	¥12,089
Qwen3-MoE-Rerank-0.6B	qwen3-moe-rerank-0.6b	MU5 x 1	¥21	¥10,139
Qwen3-Rerank-0.6B	qwen3-rerank-0.6b	MU5 x 1	¥21	¥10,139
Qwen3-Rerank-0.6B	qwen3-rerank-0.6b	MU6 x 1	¥25	¥12,089
Qwen3-Max-2025-09-23	qwen3-max-2025-09-23	MU2 x 8	¥504	¥240,288
Qwen3-Max-2025-09-23	qwen3-max-2025-09-23	MU3 x 8	¥1,096	¥527,752
Qwen3-Rerank	qwen3-rerank	MU5 x 1	¥21	¥10,139

Qwen2.5-Instruct-72B	qwen2.5-72b-instruct	MU1 x 4	¥216	¥104,472
Qwen2.5-Instruct-32B	qwen2.5-32b-instruct	MU1 x 4	¥216	¥104,472
Qwen2.5-Instruct-14B	qwen2.5-14b-instruct	MU1 x 2	¥108	¥52,236
Qwen2.5-Instruct-7B	qwen2.5-7b-instruct	MU1 x 2	¥108	¥52,236
Qwen2.5-Instruct-7B	qwen2.5-7b-instruct	MU5 x 1	¥21	¥10,139
Qwen2.5-Instruct-3B	qwen2.5-3b-instruct	MU5 x 1	¥21	¥10,139

Qwen-Flash-2025-07-28	qwen-flash-2025-07-28	MU1 x 4	¥216	¥104,472
Qwen-Plus-2025-07-28	qwen-plus-2025-07-28	MU1 x 4 MU1 x 16 (PD-separated mode)	¥216 ¥864	¥104,472 ¥417,888
Qwen-Plus-2025-12-01	qwen-plus-2025-12-01	MU1 x 4	¥216	¥104,472

GLM

Model name	Model code	Unit specification	Hourly price (CNY)	Monthly price (CNY)
GLM-5	glm-5	MU3 x 16 (PD separation mode)	¥2,192	¥1,055,504
GLM-4.7	glm-4.7	MU6 x 32 (PD separation mode)	¥800	¥386,848

DeepSeek

Model name	Model code	Unit configuration	Hourly rate	Monthly rate
DeepSeek-v4-Flash	deepseek-v4-flash	MU1 x 8	¥432	¥208,944
DeepSeek-v3.2	deepseek-v3.2	MU2 x 16 (pd-separated mode)	¥1,008	¥480,576

More models

Model name	Model code	Unit specification	Hourly price (CNY)	Monthly price (CNY)
MiniMax-M2.5	MiniMax-M2.5	MU1 x 16 (PD-decoupled mode)	PD-decoupled mode: ¥864	PD-decoupled mode: ¥417,888

Kimi-K2.5	Kimi-K2.5	MU2 x 8	¥504	¥240,288

graph TD A[Full model] --> B{Scenarios}; B --> B1[Small model size]; B --> B2[Simple model structure]; B --> B3[Few training epochs]; A --> C{Advantages}; C --> C1[Fast training]; C --> C2[Low resource consumption]; A --> D{Disadvantages}; D --> D1[Low model accuracy]; D --> D2[Prone to underfitting]; E[Incremental model] --> F{Scenarios}; F --> F1[Large model size]; F --> F2[Complex model structure]; F --> F3[Many training epochs]; E --> G{Advantages}; G --> G1[High model accuracy]; E --> H{Disadvantages}; H --> H1[Slow training]; H --> H2[High resource consumption];

Instruct - After deployment, the model runs inference in instruct mode.
Thinking - After deployment, the model runs inference in thinking mode.

Model deployment type:

pd-separated mode: Reduces first-token latency and improves throughput.

This deployment mode splits model inference into two phases, prefill and decode, which run on separate compute nodes.

Multimodal

Qwen-VL

Model name	Model code	Model unit	Hourly price	Monthly price
Qwen3-VL-235B-A22B-Instruct	qwen3-vl-235b-a22b-instruct	MU1 x 4	¥216	¥104,472
Qwen3-VL-235B-A22B-Thinking	qwen3-vl-235b-a22b-thinking	MU1 x 4	¥216	¥104,472
Qwen3-VL-32B-Instruct	qwen3-vl-32b-instruct	MU2 x 8	¥504	¥240,288
Qwen3-VL-8B-Instruct	qwen3-vl-8b-instruct	MU1 x 2	¥108	¥52,236
Qwen3-VL-4B-Instruct	qwen3-vl-4b-instruct	MU1 x 2	¥108	¥52,236
Qwen3-VL-2B-Instruct	qwen3-vl-2b-instruct	MU5 x 1	¥21	¥10,139
Qwen3-VL-Embedding-2B	qwen3-vl-embedding-2b	MU5 x 1	¥21	¥10,139
Qwen3-VL-Flash-2025-10-15	qwen3-vl-flash-2025-10-15	MU1 x 4	¥216	¥104,472
Qwen3-VL-Plus-2025-09-23	qwen3-vl-plus-2025-09-23	MU1 x 4	¥216	¥104,472

Qwen-VL-Max-2025-08-13	qwen-vl-max-2025-08-13	MU6 x 4	¥100	¥48,356
Qwen-VL-OCR-2025-11-20	qwen-vl-ocr-2025-11-20	MU6 x 4	¥100	¥48,356

Qwen Omni

Model name	Model code	Model unit	Hourly price	Monthly price
Qwen3.5-Omni-Flash	qwen3.5-omni-flash	MU8 x 1	¥47	¥22,400
Qwen3.5-Omni-Flash	qwen3.5-omni-flash	MU9 x 1	¥51	¥24,600
Qwen3.5-Omni-Plus	qwen3.5-omni-plus	MU9 x 8	¥408	¥196,800

Model types:

Instruct - After deployment, the model runs in non-thinking mode.
Thinking - After deployment, the model runs in thinking mode.
Instruct/Thinking - You can enable or disable thinking mode during model deployment.

Speech synthesis

CosyVoice

Model name	Model code	Model unit specification	Hourly price (CNY)	Monthly price (CNY)
cosyvoice-v3-flash	cosyvoice-v3-flash	MU5	¥21	¥10,139

Token-based billing

cost = (number of input tokens × price per input token) + (number of output tokens × price per output token) (minimum billing unit: 1 token)

Token-based billing applies only to custom models fine-tuned from the following foundation models.

Qwen

Foundation model	Model code	Input CNY/1,000 tokens	Output CNY/1,000 tokens
Qwen3-32B	qwen3-32b	¥0.002	non-thinking mode: ¥0.008 thinking mode: ¥0.02
Qwen3-14B	qwen3-14b	¥0.001	non-thinking mode: ¥0.004 thinking mode: ¥0.01
Qwen3-8B	qwen3-8b	¥0.0005	non-thinking mode: ¥0.002 thinking mode: ¥0.005

Qwen2.5-72B-Instruct	qwen2.5-72b-instruct	¥0.004	¥0.012
Qwen2.5-32B-Instruct	qwen2.5-32b-instruct	¥0.002	¥0.006
Qwen2.5-14B-Instruct	qwen2.5-14b-instruct	¥0.001	¥0.003
Qwen2.5-7B-Instruct	qwen2.5-7b-instruct	¥0.0005	¥0.001

Qwen-VL

Foundation model	Model code	Input CNY/1,000 tokens	Output CNY/1,000 tokens
Qwen3-VL-8B-Instruct	qwen3-vl-8b-instruct	¥0.0005	¥0.002

Qwen2.5-VL-72B-Instruct	qwen2.5-vl-72b-instruct	¥0.016	¥0.048
Qwen2.5-VL-32B-Instruct	qwen2.5-vl-32b-instruct	¥0.008	¥0.024
Qwen2.5-VL-7B-Instruct	qwen2.5-vl-7b-instruct	¥0.002	¥0.005

Image generation models – Wan

Deployment is free. Invocations are billed at the standard rate of the fine-tuned base model. For the training workflow, see Fine-tune image generation models.

Model ID	LoRA Deployment & Invocation Price
wan2.7-image-pro	CNY 0.50/image
wan2.7-image	CNY 0.20/image

FAQ

Q: When does billing for model deployment start?

A: Billing starts when the model status changes to Running. No charges apply during Deploying, Overdue Payment, or Deployment Failed.

For monthly subscriptions, the billing period starts when the status changes to Running.

Q: Am I charged if I cancel a training job?

A: Yes. If you cancel training manually, you are charged for all tokens processed before cancellation. Training jobs interrupted by system errors or other non-user causes are not charged.

Q: How do I view invocation statistics for a deployed model?

A: Visit the Model Monitoring (Beijing), Model Monitoring (Virginia), or Model Monitoring (Singapore) page.