Model Marketplace

The Model Marketplace is the primary entry point for discovering and deploying models. Models published by platform administrators are displayed here as cards, with support for multi-dimensional filtering, performance references, pricing comparison, and one-click try / one-click deploy.

Access path: Left sidebar ModelX Service > Model Marketplace, page route /admin/model-x/models.

Browsing and Filtering Models

The left side of the page is the filter panel; the right side is the model card list:

1. Category filter: Click a model category on the left (e.g., LLM, Embedding, Vision, Rerank, Audio).
2. Tag filter: Filter by model scenario, model size, context length, inference engine, chip vendor, model provider, and other tags.
3. Keyword search: Enter a model name in the search box at the top for real-time filtering.
4. Available for try toggle: Turn on to show only models that support online playground.

Each model card displays:

• Model name and release date
• Model tags
• Pricing: Token-based billing shows input / output unit price; per-call billing shows price per call
• Average latency and error rate over the past 24 hours (as a reference for runtime quality)
• Model provider icon

One-Click Deploy

1. Locate the target model card on the Model Marketplace.
2. Click "Deploy" (only shown for models that support local deployment).
3. The system automatically navigates to the inference service creation page with the following pre-filled fields:
   • Model source: Model Marketplace
   • Model: current model
   • Deployment template: the default template associated with the model
4. Supplement additional fields such as project, compute specification, and instance count.
5. Click "Create" and wait for the service to start running to obtain the API call endpoint.

One-Click Try

1. Locate a model card that supports online playground on the Model Marketplace.
2. Click "Try Now".
3. The system navigates to ModelX Service > Online Playground and automatically selects that model.
4. Enter content in the chat / image generation / voice pages to begin testing.

Model Details

Click a model name to enter the details page:

• Basic info: Model category, context length, maximum output length, parameter count, model provider, tags.
• Pricing section: Displays input / output / cache-read / cache-write-5m / cache-write-1h pricing tiers.
• Performance reference: Latency, error rate, and call volume trends over the past 24 hours.
• Deployment records: List of inference services associated with the model (if authorized).

FAQ

Q: Why don't some models have a "Deploy" button? A: The administrator did not enable "Support local deployment" when creating the model, or the model is only available for external channel calls.

Q: Where do the pricing and performance data come from? A: Pricing comes from billing policies configured by administrators; performance data is based on real call statistics from the platform over the past 24 hours and is for reference only.