A MaxCompute model is a deployment object that you register in MaxCompute and use for prediction or generation tasks. Models bring large language model (LLM) and machine learning capabilities into the same platform where your data lives — no data movement required.
Key concepts
| Term | Definition |
|---|---|
| Model | A deployment object registered in MaxCompute. Call it through AI Functions to run prediction or generation tasks directly on your data. |
| Model version | An independent, uniquely identifiable sub-object under a model. Create and manage multiple iterative versions under the same model name to support grayscale releases, rapid rollbacks, and side-by-side performance comparisons. |
What you can do with models
Register and version models with the same permission controls you use for data.
Call models from SQL or Python (MaxFrame) without exporting data to an external system.
Run AI inference where your data already lives, eliminating the security risks, costs, and latency of data movement.
Model types
MaxCompute provides four model types. The table below shows which management interfaces each type supports.
| Model type | Description | Supported management | Tutorial |
|---|---|---|---|
| Public model | Built-in, open-source large models hosted in the BIGDATA_PUBLIC_MODELSET project. Ready to call immediately through AI Functions — no registration required. | Console (view only) | Use a MaxCompute public model for sentiment analysis of online reviews |
| Remote model | A model already deployed on PAI-EAS, registered in MaxCompute by specifying a PAI-EAS Endpoint and token. | SQL, console (view only) | Use a MaxCompute remote model to automatically generate E-commerce product descriptions |
| Internally trained model | A traditional machine learning model trained with MaxFrame and saved to MaxCompute using TO_ODPS_MODEL. | SQL, MaxFrame (create only), console (view only) | Use MaxCompute for XGBoost model training and prediction |
| Imported model | A custom model trained and fine-tuned externally, imported into MaxCompute by specifying its OSS (Object Storage Service) path. | Being rolled out | Being rolled out |
Public models
The following models are available in the BIGDATA_PUBLIC_MODELSET project, in the schema named default:
Qwen3-0.6B-GGUF
Qwen3-1.7B-GGUF
Qwen3-4B-GGUF
Qwen3-8B-GGUF
Qwen3-14B-GGUF
DeepSeek-R1-Distill-Qwen-1.5B
DeepSeek-R1-Distill-Qwen-7B
DeepSeek-R1-Distill-Qwen-14B
DeepSeek-R1-0528-Qwen3-8B
To reference a public model, use the format that matches your project's schema syntax setting:
| Schema syntax setting | Reference format |
|---|---|
| Not enabled | bigdata_public_modelset.<model_name> |
| Enabled | bigdata_public_modelset.default.<model_name> |
Public models are available in the following regions: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Ulanqab), China (Hong Kong), Singapore, Indonesia (Jakarta), and Germany (Frankfurt). Support for other regions is being rolled out.
Manage models
Before managing models, make sure your account has the permissions to manage model objects.
Three management interfaces are available:
| Interface | Capabilities | Instructions |
|---|---|---|
| SQL | Create, view, modify, and delete models | Create and manage models using SQL |
| MaxFrame | Create models (Python) | Manage models using MaxFrame |
| Console | View models (graphical user interface) | See steps below |
Console model management is currently available only in the China (Beijing) region. Support for other regions is being rolled out.
To view models in the console:
Log on to the MaxCompute console and select a region in the top-left corner.
In the left navigation pane, choose .
On the Projects page, find the target project and click Manage in its Actions column.
On the Project Settings page, click the Models tab.
The Models tab shows public models from BIGDATA_PUBLIC_MODELSET and any other model types you have created, along with their version information.