Skip to main content
Watch: I Built an AI App Without Leaving Databricks

Fine-tuning - verified

Alibaba Model Studio fine-tuning: status, methods and prices

Offered. 3 offerings recorded: SFT, CPT and DPO. Training is priced per 1,000 training tokens, in Alibaba Cloud Model Studio's own units. Weight export: No.

Recorded as: Alibaba Cloud Model Studio.

Read from alibabacloud.com. Compare every fine-tuning API

Status
Offered
Offerings
3
Methods
SFT, CPT, DPO
Export weights
No

Status

Is Alibaba Model Studio fine-tuning still available?

Offered

Offered

All 3 offerings

Prices are in the provider's own unit, as written. Not stated means the page we checked does not say.
SFTNot stated

SFT (full-parameter and efficient/LoRA), Singapore

Supervised fine-tuning

Status
Not stated
Tunable base models
  • qwen3.8-27b
  • qwen3.6-27b
  • qwen3.5-27b
  • qwen3.5-9b
  • qwen3-14b (LoRA only)
“Qwen3.8-27B · qwen3.8-27b · ✓ · ✓”
alibabacloud.com
“Qwen3.6-27B · qwen3.6-27b · ✓ · ✓”
alibabacloud.com
“Qwen3.5-27B · qwen3.5-27b · ✓ · ✓”
alibabacloud.com
“Qwen3.5-9B · qwen3.5-9b · ✓ · ✓”
alibabacloud.com
“Qwen3-14B · qwen3-14b · × · ✓”
alibabacloud.com
Training price
SFT (full-parameter and efficient/LoRA), Singapore training prices in Alibaba Cloud Model Studio's own units
ModelPriceUnitQuote
qwen3.8-27b$0.0075per 1,000 training tokens (page unit: 'Thousand Token'; Singapore region)“Qwen3.8-27B · qwen3.8-27b · $0.0075/Thousand Token”alibabacloud.com
qwen3.6-27b$0.0075per 1,000 training tokens (page unit: 'Thousand Token'; Singapore region)“Qwen3.6-27B · qwen3.6-27b · $0.0075/Thousand Token”alibabacloud.com
qwen3.5-27b$0.0075per 1,000 training tokens (page unit: 'Thousand Token'; Singapore region)“Qwen3.5-27B · qwen3.5-27b · $0.0075/Thousand Token”alibabacloud.com
qwen3.5-9b$0.003per 1,000 training tokens (page unit: 'Thousand Token'; Singapore region)“Qwen3.5-9B · qwen3.5-9b · $0.003/Thousand Token”alibabacloud.com
qwen3-14b$0.0016per 1,000 training tokens (page unit: 'Thousand Token'; Singapore region)“Qwen3-14B · qwen3-14b · $0.0016/Thousand Token”alibabacloud.com
Tuned-model inference
Not stated
Hosting requirement

dedicated deployment required before use

“A fine-tuned model can only be used after deployment”
alibabacloud.com
Export weights

No

“Exported parameter snapshots are saved in cloud storage and cannot be accessed or downloaded.”
alibabacloud.com
Data limits

  • SFT needs at least 1,000 entries (recommended)

    “For SFT, the dataset requires at least 1,000 high-quality fine-tuning data entries.”
    alibabacloud.com
Notes

Region restriction: 'only available in the Singapore region' (fine-tuning API page).

CPTNot stated

Continual pre-training (CPT)

Continued pre-training

InferredThese model lists come from the table under the second region tab of the overview. We attribute that table to North China 2 (Beijing) by tab order; the page text does not name the region beside it.

Status
Not stated
Tunable base models
  • qwen3.6-plus-2026-04-02
  • qwen3-32b
  • qwen3-30b-a3b-instruct-2507
  • qwen3-4b-instruct-2507
  • qwen3-1.7b
  • qwen3-0.6b
  • qwen2.5-72b-instruct
  • qwen2.5-32b-instruct
  • qwen2.5-14b-instruct
  • qwen2.5-7b-instruct

From the table under the second region tab (Beijing by tab order, inferred).

“Qwen3.6-Plus-2026-04-02 · qwen3.6-plus-2026-04-02 · ✓ · ✓ · × · × · ×”
alibabacloud.com
“Qwen3-32B · qwen3-32b · ✓ · ✓ · ✓ · ✓ · ✓”
alibabacloud.com
Training price

Not stated

Price for this region tab did not render in the fetched text; not available in the Singapore tab.

Tuned-model inference
Not stated
Hosting requirement

dedicated deployment required before use

“A fine-tuned model can only be used after deployment”
alibabacloud.com
Export weights

No

“Exported parameter snapshots are saved in cloud storage and cannot be accessed or downloaded.”
alibabacloud.com
Data limits

  • CPT needs at least 50 million tokens (recommended)

    “For CPT, the dataset requires at least 50 million tokens of high-quality pre-training data.”
    alibabacloud.com
DPONot stated

Direct preference optimization (DPO)

Direct preference optimization (preference tuning)

InferredThese model lists come from the table under the second region tab of the overview. We attribute that table to North China 2 (Beijing) by tab order; the page text does not name the region beside it.

Status
Not stated
Tunable base models
  • qwen3-32b
  • qwen3-14b
  • qwen3-8b
  • qwen3-4b-instruct-2507
  • qwen3-1.7b
  • qwen3-0.6b
  • qwen2.5-72b-instruct
  • qwen2.5-32b-instruct
  • qwen2.5-14b-instruct
  • qwen2.5-7b-instruct
  • qwen-plus-character-2025-11-06

From the table under the second region tab (Beijing by tab order, inferred).

“Qwen3-14B · qwen3-14b · × · ✓ · ✓ · ✓ · ✓”
alibabacloud.com
“Qwen-Plus-Character-2025-11-06 · qwen-plus-character-2025-11-06 · × · ✓ · ✓ · ✓ · ✓”
alibabacloud.com
Training price

Not stated

Not offered in the Singapore tab; price not captured.

Tuned-model inference
Not stated
Hosting requirement

dedicated deployment required before use

“A fine-tuned model can only be used after deployment”
alibabacloud.com
Export weights

No

“Exported parameter snapshots are saved in cloud storage and cannot be accessed or downloaded.”
alibabacloud.com
Data limits
Not stated

Research notes

Fine-tuning API is only available in the Singapore region for the international site. The overview has region tabs (Singapore, North China 2 Beijing); the Singapore tab lists SFT only for five Qwen models. A second, larger table with CPT/SFT/DPO columns appears under the second tab and is attributed to Beijing by tab order (inferred). Billing = (training tokens + mixed training tokens) x epochs x unit price.

Plain language

The methods, explained

SFTSupervised fine-tuning
You show the model example prompts with the answers you want, and it learns to answer like that. The default method almost everywhere.
DPODirect preference optimization (preference tuning)
You give pairs of answers to the same prompt, one preferred and one not, and the model learns to lean toward the preferred kind. Good for tone and style.
RFTReinforcement fine-tuning
The model attempts tasks and a grader (your code, or another model) scores each attempt; training pushes it toward higher-scoring answers. Often billed by training hour, plus the grader's tokens.
Vision SFTVision fine-tuning
Supervised fine-tuning where the examples include images, so the model gets better at a visual task.
DistillationDistillation
A large teacher model writes answers that are used to fine-tune a smaller, cheaper student model. You pay for the teacher's calls as well as the training.
LoRALoRA (adapter) fine-tuning
Trains a small set of extra weights, an adapter, on top of the frozen base model. Cheaper and faster than full fine-tuning; the adapter is served alongside the base model.
FullFull-parameter fine-tuning
Updates every weight in the model. Costs more than LoRA and produces a complete copy of the model.
CPTContinued pre-training
Keeps training the model on a large body of raw, unlabeled text from your domain, so it absorbs vocabulary and facts before any instruction tuning.

FAQ

Does Alibaba Cloud Model Studio offer fine-tuning?

Offered. 3 offerings: SFT, CPT and DPO.

How much does Alibaba Cloud Model Studio fine-tuning cost?

Per 1,000 training tokens: qwen3.8-27b $0.0075, qwen3.6-27b $0.0075, qwen3.5-27b $0.0075, qwen3.5-9b $0.003 and 1 more. Not stated for Continual pre-training (CPT) and Direct preference optimization (DPO). Prices are in Alibaba Cloud Model Studio's own units and are not converted.

Can you download a model you fine-tune on Alibaba Cloud Model Studio?

No. The docs say: "Exported parameter snapshots are saved in cloud storage and cannot be accessed or downloaded."

Sources

Last verified . This is a directory of what each provider's own pages say, not a quality ranking. Prices are copied in the provider's own unit and never converted, so rows in different units are not comparable.

Newsletter

Know when a provider changes its fine-tuning terms

One email per week. New model releases, deprecations and pricing changes, and what they mean for what you build.