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

NoteFor the training workflow, see Introduction to model fine-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)

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

Qwen-VL

Model service	Model Code	Price
Qwen3-VL-8B-Instruct	qwen3-vl-8b-instruct	¥0.012/1K tokens
Qwen3-VL-8B-Thinking	qwen3-vl-8b-thinking	¥0.012/1K tokens
Qwen3-VL-4B-Instruct	qwen3-vl-4b-instruct	¥0.006/1K tokens

Qwen2.5-VL-72B-Instruct	qwen2.5-vl-72b-instruct	¥0.05/1K tokens
Qwen2.5-VL-32B-Instruct	qwen2.5-vl-32b-instruct	¥0.02/1K tokens
Qwen2.5-VL-7B-Instruct	qwen2.5-vl-7b-instruct	¥0.01/1K tokens

Qwen

Model service	Model Code	Price
Qwen3.8-27B	qwen3.8-27b	¥0.05/1K tokens

Qwen3.7-Plus-2026-05-26	qwen3.7-plus-2026-05-26	¥0.35/1K tokens

Qwen3.6-27B	qwen3.6-27b	¥0.05/1K tokens
Qwen3.6-Flash-2026-04-16	qwen3.6-flash-2026-04-16	¥0.05/1K tokens
Qwen3.6-Plus-2026-04-02	qwen3.6-plus-2026-04-02	¥0.3/1K tokens

Qwen3.5-27B	qwen3.5-27b	¥0.05/1K tokens
Qwen3.5-9B	qwen3.5-9b	¥0.02/1K tokens
Qwen3.5-4B	qwen3.5-4b	¥0.015/1K tokens
Qwen3.5-Flash-2026-02-23	qwen3.5-flash-2026-02-23	¥0.05/1K tokens
Qwen3.5-Plus-2026-02-15	qwen3.5-plus-2026-02-15	¥0.3/1K tokens

Qwen3-32B	qwen3-32b	¥0.04/1K tokens
Qwen3-30B-A3B-Instruct-2507	qwen3-30b-a3b-instruct-2507	¥0.03/1K tokens
Qwen3-14B	qwen3-14b	¥0.03/1K tokens
Qwen3-8B	qwen3-8b	¥0.006/1K tokens
Qwen3-4B-Instruct-2507	qwen3-4b-instruct-2507	¥0.006/1K tokens
Qwen3-1.7B	qwen3-1.7b	¥0.0045/1K tokens
Qwen3-0.6B	qwen3-0.6b	¥0.003/1K tokens

Qwen2.5-72B-Instruct	qwen2.5-72b-instruct	¥0.15/1K tokens
Qwen2.5-32B-Instruct	qwen2.5-32b-instruct	¥0.03/1K tokens
Qwen2.5-14B-Instruct	qwen2.5-14b-instruct	¥0.03/1K tokens
Qwen2.5-7B-Instruct	qwen2.5-7b-instruct	¥0.006/1K tokens

Qwen-Plus-Character-2025-11-06	qwen-plus-character-2025-11-06	¥0.15/1K tokens

Image generation models – Wan

NoteFor 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 1K tokens)

Formula for total training tokens

`Training Tokens Total\approx max\_steps \times L_{step}`

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:
$`L_{step} = \sum_{i \in batch} L_{item}^{(i)} \leq L_{max}`$
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	12,800
t2i (text-to-image)	2k	23,220
i2i (image-to-image)	1k	23,220
i2i (image-to-image)	2k	32,000

Model service	Model Name	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 = 12,800 (generation_type=t2i, max_token_length=1k), L_step ≈ L_max = 12,800
Total training tokens ≈ 200 × 12800 = 2560000 = 2560 thousand tokens
Model training fee ≈ 2560 × 0.08 = CNY 204.8

Image generation models – Qwen

NoteFor 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 1K tokens)

Formula for total training tokens

`\text{Total training tokens} = n\_epochs \times N \times (max\_pixels / \text{Compression ratio}) \times \text{GPU coefficient}`

Where:

n_epochs: A hyperparameter specified during training, representing the number of training epochs (configured when creating a fine-tuning job).
N: The total number of images in the training set.
max_pixels: A hyperparameter specified during training, representing the maximum number of pixels per image (configured when creating a fine-tuning job). Valid values are 1k (1024×1024) and 2k (2048×2048).
Compression ratio: The fixed VAE compression ratio, which is 1024 (32×32).
GPU coefficient: A resource scheduling coefficient that is dynamically adjusted based on job scheduling.

Model service	Model Name	Training price (per 1K tokens)
Qwen image generation	qwen-image-2.0	CNY 0.02
Qwen image generation	qwen-image-2.0-pro	CNY 0.02

Billing example

Suppose you fine-tune the qwen-image-2.0 model with a training set of 1 image. The parameters are: n_epochs = 1000, max_pixels = "1k", the training price is CNY 0.02 per 1,000 tokens, and the GPU coefficient is 8:

Tokens per image = max_pixels / Compression ratio = 1024×1024 / 1024 = 1024
Total training tokens = 1000 × 1 × 1024 × 8 = 8192000 = 8192 thousand tokens
Model training fee = 8192 × 0.02 = CNY 163.84

The following table estimates token consumption and fees per training image, assuming a GPU coefficient of 8. For a training set of N images, multiply the values by N.

max_pixels	n_epochs	Estimated tokens	Estimated cost (CNY)
1k	800	6,553,600	131.07
1k	1,000	8,192,000	163.84
1k	2,000	16,384,000	327.68
2k	800	26,214,400	524.29
2k	1,000	32,768,000	655.36
2k	2,000	65,536,000	1310.72

Note

The GPU coefficient is dynamically adjusted based on job scheduling. The value 8 in the example and the table above is used only to illustrate the calculation. For actual token consumption, see the usage field returned by the Query a fine-tuning job operation. Your bill is the final authority on fees.
batch_size does not affect billing or the total training tokens.

Video generation models – Wan

NoteFor 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 1K tokens)

Formula for total training tokens

`Training Tokens Total = \left( \sum_{i=1}^{N} \text{billing duration of video}_i \right) \times \frac{max\_pixels}{1024} \times n\_epochs`

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).
- The conversion between n_epochs and steps is: steps = n_epochs × ⌈dataset_size / batch_size⌉, i.e., n_epochs = steps / ⌈dataset_size / batch_size⌉.
- When the dataset contains only 1 sample and batch_size = 1, n_epochs = steps. We recommend a total of at least 800 steps.
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.
- wan2.7 model: Billing duration=min(10, rounded duration), meaning a single video is billed for a maximum of 10 seconds.
- 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 service	Model Name	Training price (per 1K tokens)
Wan image-to-video (first frame-based)	wan2.7-i2v	CNY 2
	wan2.2-i2v-flash	CNY 0.06
	wan2.5-i2v-preview	CNY 0.32
Image-to-video (first and last frame-based)	wan2.7-i2v	CNY 2
Image-to-video (first and last frame-based)	wan2.2-kf2v-flash	CNY 0.06

Billing example

wan2.7-i2v cost estimation (single data)

Assume a training set contains 1 video with a duration of 10 seconds. With batch_size = 1 (recommended), n_epochs = steps / ⌈1(dataset_size) / 1(batch_size)⌉ = steps.

Training unit price = CNY 2/thousand tokens. Taking max_pixels = 36864 and n_epochs = 800 as an example:

Total training tokens = 10 × (36864 / 1024) × 800 = 288,000 = 288 thousand tokens
Model training fee = 288 × 2 = CNY 576

max_pixels	Common steps	n_epochs	Estimated tokens	Estimated cost (CNY)
36864	800	800	288,000	576
	1,000	1,000	360,000	720
	2,000	2,000	720,000	1,440
65536	800	800	512,000	1,024
	1,000	1,000	640,000	1,280
	2,000	2,000	1,280,000	2,560
102400	800	800	800,000	1,600
	1,000	1,000	1,000,000	2,000
	2,000	2,000	2,000,000	4,000

wan2.7-i2v cost estimation (multiple data)

Assume a training set contains 2 videos with durations of 3.4 seconds and 11.5 seconds. Parameters: max_pixels = 36864, n_epochs = 800. Training unit price = CNY 2/thousand tokens:

Duration calculation:
- Video 1: 3.4 seconds is rounded to 3. Billable duration = min(10, 3) = 3.
- Video 2: 11.5 seconds is rounded to 12. Billable duration = min(10, 12) = 10.
- Total billable duration = 3 + 10 = 13 seconds.
Total training tokens = 13 × (36864/1024) × 800 = 374400 = 374.4 thousand tokens
Model training fee = 374.4 × 2 = CNY 748.8

Speech synthesis models – CosyVoice

NoteCosyVoice model fine-tuning is available only in the China (Beijing) region. For the training workflow, see Fine-tune CosyVoice. After training completes, deploy the new model before calling it.

Method	Billed by tokens consumed during training
Unit price	CNY 0.2/1K tokens
Formula	Model training fee = Total consumed tokens × Training unit price

Formula for total consumed tokens

The token consumption of a single job is estimated as follows:

`\text{Consumed tokens} = (lm\_max\_epoch + fm\_max\_epoch) \times 25 \times \text{Training set duration (s)}`

Where lm_max_epoch and fm_max_epoch are hyperparameters set when creating the fine-tuning job, representing the number of LM and FM training epochs. The training set duration is the total duration in seconds of all audio files in the fine-tuning dataset. Increasing either epoch count or enlarging the training set increases token consumption linearly.

Deployment billing

Text generation models – Qwen

Billing by usage duration (Provisioned Throughput)

Fee = Usage duration × (Input TPM unit price × Input TPM + Output TPM unit price × Output TPM)

Post-paid is calculated hourly: the usage duration unit is hours, and the unit price is taken from the "Continuous 1 hour" column in the table below; prepaid is calculated daily: the usage duration unit is days, and the unit price is taken from the "Continuous 1 day" column in the table below.

Prepaid orders take effect in real time after payment, with a validity period of N days ending at 23:59 on day N. If the order is placed after 22:00, the expiration date will be automatically extended by 1 day.
After a prepaid order expires, the service will be stopped with a 2-hour delay, and resources will be retained for 14 hours after the stop and then released.
Prepaid orders support early termination; the used portion is settled at a 1.2x coefficient for refund. For details, see Refund rules for configuration downgrades.
For post-paid billing, if the account is in arrears, the deployed resources will become unavailable and will no longer be billed. After the arrears are paid, the service will resume.

When the model input exceeds the maximum input Token, the relevant call will automatically switch to the pay-as-you-go mode of the current model; when the purchased TPM is exceeded, it is handled according to the overflow strategy selected at creation ("auto-overflow" switches to pay-as-you-go, "use-only-PTU-capacity" returns 429). At this time, inference performance may degrade and will be subject to the public traffic control of the current snapshot model in the business space, and fees will be charged according to the model invocation (pay-as-you-go) standard.

In this case (only under the "auto-overflow" strategy), the API response Header will include: x-dashscope-ptu-overflow:true.
For TPM statistics, go to: Model Monitoring (Beijing).

For the specific fee reduction and refund rules in scale-down (downgrade) scenarios, please refer to: Refund rules for configuration downgrades.

NotePTU deployment supports long-input tiered capacity coefficients and cache discounts; see Provisioned Throughput long input and caching for details.

North China 2 (Beijing)

Qwen

Model Name	Model Code	Max input tokens	Post-paid input Per 10K TPM/hour	Post-paid output Per 1K TPM/hour	Prepaid input Per 10K TPM/day	Prepaid output Per 1K TPM/day
Qwen3.8-Flash	qwen3.8-flash	128K	¥2.88	¥0.97	¥34.56	¥11.66
Qwen3.8-Max	qwen3.8-max	1M	¥28.8	¥8.64	¥345.6	¥103.68

Qwen3.7-Flash-2026-07-15	qwen3.7-flash-2026-07-15	128K	¥0.48	¥0.19	¥5.76	¥2.3
Qwen3.7-Max-2026-05-20	qwen3.7-max-2026-05-20	256K	¥28.8	¥8.64	¥345.6	¥103.68
Qwen3.7-Plus-2026-05-26	qwen3.7-plus-2026-05-26	256K	¥4.8	¥1.92	¥57.6	¥23.04

Qwen3.6-Plus-2026-04-02	qwen3.6-plus-2026-04-02	128K	¥4.8	¥2.88	¥57.6	¥34.56

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

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

Qwen-Flash-2025-07-28	qwen-flash-2025-07-28	128K	¥0.36	¥0.36	¥4.32	¥4.32
Qwen-Plus-2025-12-01	qwen-plus-2025-12-01	128K	¥1.92	Non-thinking: ¥0.48 Thinking: ¥1.92	¥23.04	Non-thinking: ¥5.76 Thinking: ¥23.04

DeepSeek

Model Name	Model Code	Max input tokens	Post-paid input Per 10K TPM/hour	Post-paid output Per 1K TPM/hour	Prepaid input Per 10K TPM/day	Prepaid output Per 1K TPM/day
DeepSeek-v4-Flash	deepseek-v4-flash	256K	¥3.6	¥0.72	¥43.2	¥8.64
DeepSeek-v4-Flash-0731	deepseek-v4-flash-0731	256K	¥7.2	¥1.44	¥86.4	¥17.28
DeepSeek-v4-Pro	deepseek-v4-pro	256K	¥43.2	¥8.64	¥518.4	¥103.68
DeepSeek-v3.2	deepseek-v3.2	64K	¥7.2	¥1.08	¥86.4	¥12.96
DeepSeek-v3	deepseek-v3	64K	¥7.2	¥2.88	¥86.4	¥34.56

GLM

Model Name	Model Code	Max input tokens	Post-paid input Per 10K TPM/hour	Post-paid output Per 1K TPM/hour	Prepaid input Per 10K TPM/day	Prepaid output Per 1K TPM/day
GLM-5.2	glm-5.2	1M	¥28.8	¥10.08	¥345.6	¥120.96
GLM-5.1	glm-5.1	64K	¥21.6	¥8.64	¥259.2	¥103.68

Model Name

Model Code

Max input tokens

Post-paid input

Per 10K TPM/hour

Post-paid output

Per 1K TPM/hour

Prepaid input

Per 10K TPM/day

Prepaid output

Per 1K TPM/day

GLM-5.2

glm-5.2

¥28.8

¥10.08

¥345.6

¥120.96

GLM-5.1

glm-5.1

64K

¥21.6

¥8.64

¥259.2

¥103.68

Qwen-VL

Model Name	Model Code	Max input tokens	Post-paid input Per 10K TPM/hour	Post-paid output Per 1K TPM/hour	Prepaid input Per 10K TPM/day	Prepaid output Per 1K TPM/day
Qwen3-VL-Plus-2025-09-23	qwen3-vl-plus-2025-09-23	128K	¥2.4	¥2.4	¥28.8	¥28.8

Model Name

Model Code

Max input tokens

Post-paid input

Per 10K TPM/hour

Post-paid output

Per 1K TPM/hour

Prepaid input

Per 10K TPM/day

Prepaid output

Per 1K TPM/day

Qwen3-VL-Plus-2025-09-23

qwen3-vl-plus-2025-09-23

128K

¥2.4

¥28.8

Singapore

Qwen

Model Name	Model Code	Max input tokens	Post-paid input Per 10K TPM/hour	Post-paid output Per 1K TPM/hour	Prepaid input Per 10K TPM/day	Prepaid output Per 1K TPM/day
Qwen3.8-Flash	qwen3.8-flash	128K	¥3.94	¥1.23	¥47.25	¥14.81
Qwen3.8-Max	qwen3.8-max	1M	¥35.97	¥10.79	¥431.7	¥129.5

Qwen3.7-Flash-2026-07-15	qwen3.7-flash-2026-07-15	128K	¥0.54	¥0.23	¥6.47	¥2.81
Qwen3.7-Max-2026-05-20	qwen3.7-max-2026-05-20	256K	¥44.97	¥13.49	¥539.6	¥161.87
Qwen3.7-Plus-2026-05-26	qwen3.7-plus-2026-05-26	256K	¥7.19	¥2.88	¥86.3	¥34.53

Qwen3.6-Plus-2026-04-02	qwen3.6-plus-2026-04-02	128K	¥9	¥5.4	¥107.9	¥64.75

Qwen3.5-Plus-2026-04-20	qwen3.5-plus-2026-04-20	128K	¥7.2	¥4.32	¥86.3	¥51.8

DeepSeek

Model Name	Model Code	Max input tokens	Post-paid input Per 10K TPM/hour	Post-paid output Per 1K TPM/hour	Prepaid input Per 10K TPM/day	Prepaid output Per 1K TPM/day
DeepSeek-v4-Flash	deepseek-v4-flash	256K	¥5.4	¥1.08	¥64.8	¥12.95
DeepSeek-v4-Flash-0731	deepseek-v4-flash-0731	256K	¥10.79	¥2.16	¥129.5	¥25.9
DeepSeek-v4-Pro	deepseek-v4-pro	256K	¥64.75	¥12.95	¥777	¥155.4
DeepSeek-v3.2	deepseek-v3.2	64K	¥15.4	¥4.61	¥184.5	¥55.36

GLM

Model Name	Model Code	Max input tokens	Post-paid input Per 10K TPM/hour	Post-paid output Per 1K TPM/hour	Prepaid input Per 10K TPM/day	Prepaid output Per 1K TPM/day
GLM-5.2	glm-5.2	1M	¥37.8	¥11.87	¥453.3	¥142.45
GLM-5.1	glm-5.1	64K	¥37.77	¥11.87	¥453.3	¥142.45

Model Name

Model Code

Max input tokens

Post-paid input

Per 10K TPM/hour

Post-paid output

Per 1K TPM/hour

Prepaid input

Per 10K TPM/day

Prepaid output

Per 1K TPM/day

GLM-5.2

glm-5.2

¥37.8

¥11.87

¥453.3

¥142.45

GLM-5.1

glm-5.1

64K

¥37.77

¥11.87

¥453.3

¥142.45

Qwen-VL

Model Name	Model Code	Max input tokens	Post-paid input Per 10K TPM/hour	Post-paid output Per 1K TPM/hour	Prepaid input Per 10K TPM/day	Prepaid output Per 1K TPM/day
Qwen3-VL-Plus-2025-09-23	qwen3-vl-plus-2025-09-23	128K	¥3.6	¥2.88	¥43.2	¥34.53

Model Name

Model Code

Max input tokens

Post-paid input

Per 10K TPM/hour

Post-paid output

Per 1K TPM/hour

Prepaid input

Per 10K TPM/day

Prepaid output

Per 1K TPM/day

Qwen3-VL-Plus-2025-09-23

qwen3-vl-plus-2025-09-23

128K

¥3.6

¥2.88

¥43.2

¥34.53

Billing by usage duration (Dedicated Throughput)

Prepaid orders support early termination; the used portion is settled at a 1.2x coefficient for refund. For details, see Refund rules for configuration downgrades.

North China 2 (Beijing)

Model	Type	Base input (TPM)	Base output (TPM)	Starting price (CNY/month)	Scale-out price (CNY/month)
qwen3.7-plus-2026-05-26	Fast	1,372,000	170,000	556,466	556,466
qwen3.6-27b	Standard	273,000	34,000	49,118	49,118
qwen3.5-397b-a17b	Standard	896,000	112,000	1,107,904	1,107,904
qwen3.5-122b-a10b	Fast	7,288,000	904,000	2,226,984	2,226,984
qwen3.5-35b-a3b	Standard	656,000	82,000	109,962	109,962
qwen3.5-27b	Standard	2,432,000	304,000	1,108,688	1,108,688
qwen3.5-27b	Efficient	291,000	36,000	48,996	48,996
qwen3.5-4b	Standard	1,072,000	134,000	109,478	109,478
glm-5.2	Fast	502,000	60,000	947,630	947,630
glm-5.1	Standard	256,000	32,000	504,704	504,704
deepseek-v4-flash	Standard	2,240,000	280,000	1,107,400	1,107,400
deepseek-v4-flash-0731	Fast	1,676,000	213,000	277,984	277,984

Model	Type	Input length	Output length	First token latency (ms)	Per-token latency (ms)
qwen3.7-plus-2026-05-26	Fast	16,000	2,000	2,418	15
qwen3.6-27b	Standard	4,000	500	1,292	19
qwen3.5-397b-a17b	Standard	4,000	500	996	27
qwen3.5-122b-a10b	Fast	16,000	2,000	568	8
qwen3.5-35b-a3b	Standard	4,000	500	471	10
qwen3.5-27b	Standard	4,000	500	703	14
qwen3.5-27b	Efficient	4,000	500	1,448	23
qwen3.5-4b	Standard	4,000	500	552	6
glm-5.2	Fast	16,000	2,000	1,558	15
glm-5.1	Standard	4,000	500	769	27
deepseek-v4-flash	Standard	4,000	500	651	19
deepseek-v4-flash-0731	Fast	16,000	2,000	1,079	13

By model token usage

Fee = Model input token count × Model input unit price + Model output token count × Model output unit price (minimum billing unit: 1 token)

Billing by model token usage is supported only after you complete efficient SFT training (that is, LoRA efficient fine-tuning; the plan parameter is set to lora for API deployment) on the following base models and obtain a custom model.

Beijing

Base model	Model Code	Input CNY/million tokens	Output CNY/million tokens
Qwen3.5-27B	qwen3.5-27b	<128K ¥0.6 128K-256K ¥1.8	<128K ¥4.8 128K-256K ¥14.4

Qwen3-32B	qwen3-32b	Non-thinking mode: ¥2 Thinking mode: ¥2	Non-thinking mode: ¥8 Thinking mode: ¥20
Qwen3-14B	qwen3-14b	Non-thinking mode: ¥1 Thinking mode: ¥1	Non-thinking mode: ¥4 Thinking mode: ¥10
Qwen3-8B	qwen3-8b	Non-thinking mode: ¥0.5 Thinking mode: ¥0.5	Non-thinking mode: ¥2 Thinking mode: ¥5

Qwen3-VL-8B-Instruct	qwen3-vl-8b-instruct	¥0.5	¥2

Qwen3-4B-Instruct-2507	qwen3-4b-instruct-2507	Non-thinking mode: ¥0.3 Thinking mode: ¥0.3	Non-thinking mode: ¥1.2 Thinking mode: ¥3

Qwen2.5-Open-Source-72B	qwen2.5-72b-instruct	¥4	¥12

Qwen2.5-VL-72B	qwen2.5-vl-72b-instruct	¥16	¥48

Qwen2.5-Open-Source-32B	qwen2.5-32b-instruct	¥2	¥6
Qwen2.5-Open-Source-14B	qwen2.5-14b-instruct	¥1	¥3
Qwen2.5-Open-Source-7B	qwen2.5-7b-instruct	¥0.5	¥1

Qwen2.5-VL-7B	qwen2.5-vl-7b-instruct	¥2	¥5

Singapore

Base model	Model Code	Input CNY/million tokens	Output CNY/million tokens
Qwen3.5-27B	qwen3.5-27b	¥2.202	¥17.614

Qwen3-32B	qwen3-32b	Non-thinking mode: ¥1.174 Thinking mode: ¥1.174	Non-thinking mode: ¥4.697 Thinking mode: ¥4.697
Qwen3-14B	qwen3-14b	Non-thinking mode: ¥2.569 Thinking mode: ¥2.569	Non-thinking mode: ¥10.275 Thinking mode: ¥30.825
Qwen3-8B	qwen3-8b	Non-thinking mode: ¥1.321 Thinking mode: ¥1.321	Non-thinking mode: ¥5.137 Thinking mode: ¥15.412

Qwen3-VL-8B-Instruct	qwen3-vl-8b-instruct	¥1.321	¥5.137

Qwen3-4B-Instruct-2507	qwen3-4b-instruct-2507	Non-thinking mode: ¥0.807 Thinking mode: ¥0.807	Non-thinking mode: ¥3.082 Thinking mode: ¥9.247

Qwen2.5-Open-Source-72B	qwen2.5-72b-instruct	¥10.275	¥41.1

Qwen2.5-VL-72B	qwen2.5-vl-72b-instruct	¥20.55	¥61.65

Qwen2.5-Open-Source-32B	qwen2.5-32b-instruct	¥5.137	¥20.55
Qwen2.5-Open-Source-14B	qwen2.5-14b-instruct	¥2.569	¥10.275
Qwen2.5-Open-Source-7B	qwen2.5-7b-instruct	¥1.284	¥5.137

Qwen2.5-VL-7B	qwen2.5-vl-7b-instruct	¥2.569	¥7.706

Billing by usage duration (Model Unit)

Cost = Usage duration (hours) × Number of Model Units × Model Unit unit price

The "Model Unit unit price" takes the "Hourly unit price" column in the table below for postpaid scenarios; for monthly subscription prepaid billing, the formula becomes Number of monthly subscriptions × Number of Model Units × Monthly unit price.

For the first month of a prepaid purchase, if you cancel the subscription early within the first month, the daily unit price (≈ Monthly unit price / 30) will be billed at 1.2 times the rate (less than one day is billed as one day)

NoteThe compute resources for the Model Unit postpaid method are first-come, first-served. If the purchase fails, a full refund will be issued.

Text generation

Qwen

Model Name	Model Code	Model Unit Specification	Hourly unit price (CNY) Minimum billing: minutes	Monthly subscription unit price (CNY) Minimum billing: days
Qwen3.6-35B-A3B	qwen3.6-35b-a3b	MU1 x 8	¥432	¥208,944
		MU2 x 8	¥504	¥240,288
		MU3 x 8	¥1,096	¥527,752
		MU8 x 1	¥47	¥22,400
		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-Flash-2026-04-16	qwen3.6-flash-2026-04-16	MU3 x 8	¥1,096	¥527,752
Qwen3.6-Plus-2026-04-02	qwen3.6-plus-2026-04-02	MU1 x 8 MU1 x 16 (PD disaggregation mode)	¥432 PD disaggregation mode: ¥864	¥208,944 PD disaggregation mode: ¥417,888

Qwen3.5-397B-A17B	qwen3.5-397b-a17b	MU3 x 8 MU3 x 16 (PD disaggregation mode)	¥1,096 PD disaggregation mode: ¥2,192	¥527,752 PD disaggregation mode: ¥1,055,504
Qwen3.5-397B-A17B	qwen3.5-397b-a17b	MU6 x 16	¥400	¥193,424
Qwen3.5-122B-A10B	qwen3.5-122b-a10b	MU1 x 4	¥216	¥104,472
Qwen3.5-122B-A10B	qwen3.5-122b-a10b	MU6 x 16	¥400	¥193,424
Qwen3.5-35B-A3B	qwen3.5-35b-a3b	MU1 x 2	¥108	¥52,236
		MU2 x 8	¥504	¥240,288
		MU3 x 8	¥1,096	¥527,752
		MU9 x 1	¥51	¥24,600
Qwen3.5-27B	qwen3.5-27b	MU2 x 8	¥504	¥240,288
		MU3 x 8	¥1,096	¥527,752
		MU8 x 1	¥47	¥22,400
		MU9 x 1	¥51	¥24,600
Qwen3.5-9B	qwen3.5-9b	MU1 x 2	¥108	¥52,236
Qwen3.5-9B	qwen3.5-9b	MU2 x 8	¥504	¥240,288
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 8 MU1 x 16 (PD disaggregation mode)	¥432 PD disaggregation mode: ¥864	¥208,944 PD disaggregation mode: ¥417,888
		MU2 x 8	¥504	¥240,288
		MU3 x 8 MU3 x 16 (PD disaggregation mode)	¥1,096 PD disaggregation mode: ¥2,192	¥527,752 PD disaggregation mode: ¥1,055,504

Qwen3-235B-A22B-Instruct-2507	qwen3-235b-a22b-instruct-2507	MU1 x 4	¥216	¥104,472
		MU2 x 8	¥504	¥240,288
		MU3 x 8	¥1,096	¥527,752
Qwen3-32B	qwen3-32b	MU6 x 16	¥400	¥193,424
Qwen3-30B-A3B-Thinking-2507	qwen3-30b-a3b-thinking-2507	MU1 x 2	¥108	¥52,236
Qwen3-4B	qwen3-4b	MU1 x 2	¥108	¥52,236
Qwen3-4B	qwen3-4b	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-Open-Source-72B	qwen2.5-72b-instruct	MU1 x 8	¥432	¥208,944
Qwen2.5-Open-Source-14B	qwen2.5-14b-instruct	MU1 x 2	¥108	¥52,236
Qwen2.5-Open-Source-7B	qwen2.5-7b-instruct	MU1 x 2	¥108	¥52,236
Qwen2.5-Open-Source-7B	qwen2.5-7b-instruct	MU5 x 1	¥21	¥10,139

Qwen-Plus-2025-07-28	qwen-plus-2025-07-28	MU1 x 4	¥216	¥104,472
Qwen-Plus-2025-12-01	qwen-plus-2025-12-01	MU1 x 4	¥216	¥104,472
Qwen-Plus-Character-2025-11-06	qwen-plus-character-2025-11-06	MU1 x 4	¥216	¥104,472

GLM

Model Name	Model Code	Model Unit Specification	Hourly unit price (CNY) Minimum billing: minutes	Monthly subscription unit price (CNY) Minimum billing: days
GLM-5.1	glm-5.1	MU2 x 8	¥504	¥240,288
		MU3 x 16 (PD disaggregation mode)	PD disaggregation mode: ¥2,192	PD disaggregation mode: ¥1,055,504
		MU6 x 16	¥400	¥193,424
GLM-5	glm-5	MU3 x 16 (PD disaggregation mode)	PD disaggregation mode: ¥2,192	PD disaggregation mode: ¥1,055,504
GLM-4.7	glm-4.7	MU6 x 32 (PD disaggregation mode)	PD disaggregation mode: ¥800	PD disaggregation mode: ¥386,848

DeepSeek

Model Name	Model Code	Model Unit Specification	Hourly unit price (CNY) Minimum billing: minutes	Monthly subscription unit price (CNY) Minimum billing: days
DeepSeek-v4-Flash	deepseek-v4-flash	MU1 x 8	¥432	¥208,944
MU3 x 8	¥1,096	¥527,752
DeepSeek-v3.2	deepseek-v3.2	MU2 x 16 (PD disaggregation mode)	PD disaggregation mode: ¥1,008	PD disaggregation mode: ¥480,576

Model Name

Model Code

Model Unit Specification

Hourly unit price (CNY)

Minimum billing: minutes

Monthly subscription unit price (CNY)

Minimum billing: days

DeepSeek-v4-Flash

deepseek-v4-flash

MU1 x 8

¥432

¥208,944

MU3 x 8

¥1,096

¥527,752

DeepSeek-v3.2

deepseek-v3.2

MU2 x 16 (PD disaggregation mode)

PD disaggregation mode: ¥1,008

PD disaggregation mode: ¥480,576

More models

Model Name	Model Code	Model Unit Specification	Hourly unit price (CNY) Minimum billing: minutes	Monthly subscription unit price (CNY) Minimum billing: days
Kimi-K2.5	kimi-k2.5	MU2 x 8	¥504	¥240,288

Model Name

Model Code

Model Unit Specification

Hourly unit price (CNY)

Minimum billing: minutes

Monthly subscription unit price (CNY)

Minimum billing: days

Kimi-K2.5

kimi-k2.5

MU2 x 8

¥504

¥240,288

Model type:

Instruct - The model performs inference in non-thinking mode after deployment.
Thinking - The model performs inference in thinking mode after deployment.

Model deployment type:

PD disaggregation mode - Reduces first-token latency and increases throughput.

For models deployed in this mode, during inference the first-token computation (Prefill) and the subsequent token computation (Decode) are split across different compute nodes for execution.

Multimodal

Qwen-VL

Model Name	Model Code	Model Unit Specification	Hourly unit price (CNY) Minimum billing: minutes	Monthly subscription unit price (CNY) Minimum billing: days
Qwen3-VL-235B-A22B-Thinking	qwen3-vl-235b-a22b-thinking	MU1 x 8	¥432	¥208,944
		MU2 x 8	¥504	¥240,288
		MU3 x 8	¥1,096	¥527,752
Qwen3-VL-32B-Instruct	qwen3-vl-32b-instruct	MU2 x 8	¥504	¥240,288
Qwen3-VL-32B-Instruct	qwen3-vl-32b-instruct	MU3 x 8	¥1,096	¥527,752
Qwen3-VL-30B-A3B-Instruct	qwen3-vl-30b-a3b-instruct	MU1 x 4	¥216	¥104,472
Qwen3-VL-8B-Instruct	qwen3-vl-8b-instruct	MU1 x 2	¥108	¥52,236
Qwen3-VL-8B-Instruct	qwen3-vl-8b-instruct	MU5 x 1	¥21	¥10,139
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

QwenVL-Max-2025-08-13	qwen-vl-max-2025-08-13	MU6 x 4	¥100	¥48,356

Qwen Omni

Model Name	Model Code	Model Unit Specification	Hourly unit price (CNY) Minimum billing: minutes	Monthly subscription unit price (CNY) Minimum billing: days
Qwen3.5-Omni-Flash	qwen3.5-omni-flash	MU9 x 1	¥51	¥24,600

Model Name

Model Code

Model Unit Specification

Hourly unit price (CNY)

Minimum billing: minutes

Monthly subscription unit price (CNY)

Minimum billing: days

Qwen3.5-Omni-Flash

qwen3.5-omni-flash

MU9 x 1

¥51

¥24,600

Model type:

Instruct - The model performs inference in non-thinking mode after deployment.
Thinking - The model performs inference in thinking mode after deployment.
Instruct/Thinking - You can choose whether to enable thinking mode when deploying the model.

Speech Synthesis

CosyVoice

Model Name	Model Code	Model Unit Specification	Hourly unit price (CNY)	Monthly subscription unit price (CNY)
cosyvoice-v3-flash	cosyvoice-v3-flash	MU5	¥21	¥10,139

Image generation models – Wan

For Wan image generation models fine-tuned with SFT-LoRA, deployment is free, and invocations are billed at the standard invocation rate of the fine-tuned base model. For the model training and deployment workflow, see Fine-tune image generation models.

Model Name	LoRA deployment and invocation price
wan2.7-image-pro	CNY 0.50/image
wan2.7-image	CNY 0.20/image

Image generation models – Qwen

For Qwen image generation models fine-tuned with SFT-LoRA, deployment is free, and invocations are billed at the standard invocation rate of the fine-tuned base model. For the model training and deployment workflow, see Fine-tune image generation models.

Model Name	LoRA deployment and invocation price
qwen-image-2.0	CNY 0.20/image
qwen-image-2.0-pro	CNY 0.50/image

Video generation models – Wan

Fine-tuned Wan video generation models:

Model deployment: Free
Model invocation: Billed at the standard invocation rate of the base model used for fine-tuning. For details, see model pricing

Speech synthesis models – CosyVoice

NoteCosyVoice model fine-tuning is available only in the China (Beijing) region.

After deployment, a fine-tuned CosyVoice model is billed by the <strong>usage duration of model units</strong>. The formula is: $\text{Fee} = \text{Usage duration (hours)} \times \text{Number of model units} \times \text{Model unit price}$.

Where $\text{Number of model units} = \text{Model units per replica} \times \text{Number of replicas}$. For the available deployment templates and their model unit types, see [Deployment templates](/en/model-studio/fine-tune-speech-synthesis-model-by-api) in Fine-tune CosyVoice.

Model Name	Model Code	Model Unit Specification	Hourly unit price (CNY)	Monthly subscription unit price (CNY)
cosyvoice-v3-flash	cosyvoice-v3-flash	MU5	¥21	¥10,139

FAQ

Q: When does billing for model deployment start?

A: Billing for model deployment starts when the model deployment is complete, that is, when the model status is Running. No fees are charged when the model status is Deploying, Overdue Payment, or Deployment Failed.

For monthly prepaid billing, the monthly subscription time starts to be consumed after the model status changes to Running.

Q: Am I charged if I cancel model training?

A: Yes. After you manually cancel training, the consumed tokens will still be billed. The training fee is estimated based on the consumed tokens, and the final fee is subject to your bill. For training jobs interrupted due to other reasons, Alibaba Cloud Model Studio will not charge you any training fee.

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

A: Visit the Model Monitoring (Beijing), Model Monitoring (Virginia), Model Monitoring (Singapore), or Model Monitoring (China (Hong Kong)) page (China (Hong Kong) supports only basic monitoring; advanced monitoring and logs are unavailable).

The Model Monitoring page supports filtering by time range (3 hours, 6 hours, 12 hours, 24 hours, today, yesterday, one week, custom) and by inference type. The summary area at the top shows the total number of models, total invocations, total failures, average invocation duration, and average time to first packet, and the model list at the bottom shows metrics such as total invocations, failed invocations, failure rate, average invocation duration, and average first-token latency for each model. Click Monitor to view detailed data for a single model.