Publish an Aggregation Model

Applicable Roles

• Model Provider

Before You Start

• Prepare at least two healthy published member models with compatible protocols and modalities.
• Decide whether the aggregation objective is cost, success rate, load distribution, or high availability.
• Compare member billing, context, concurrency, and region differences.

Procedure

1. From the platform home page, select My Models in the left navigation.
2. Open the My Aggregations tab. Use Public Models / Private Models to switch between publication areas.
3. Select Create Aggregation Model in the upper-right corner.
4. Select a publication area:
   • Publish to Private Area makes the aggregation visible only within the team or tenant and does not list it in the public catalog.
   • Publish to Public Area lists the aggregation in the public catalog for all tenants and allows independent pricing and rate limits.
5. Select Publish to Public Area to open Step 1.

Step 1: Basic Information

• Select the Model Type, such as Multimodal, Chat, Image, Speech, Video, Embedding, or Rerank.
• Select the Model Subtype, such as LLM.

• Under Model Selection, select Add Model:
  • Filter the model-name and model-identifier list on the left.
  • Review provider instances on the right, including publication date, context, input and output price, throughput, success rate, weekly calls, weekly tokens, maximum output, region, and capabilities.
  • Select one or more compatible provider instances and select Confirm.

• Configure each member model:
  • Enabled controls whether the member participates in routing.
  • Minimum Success Rate, such as 80%.
  • Maximum Concurrency, such as 1500.
  • Maximum Context Length, such as 128K.
  • Input-token and output-token Cost.
  • Delete removes the member.

• Complete the aggregation's basic information:
  • Enter a Custom Identifier, such as Qwen3-235b-a22b-instruct-2507.
  • Select a Matching Policy: Cost First, Success Rate First, Balanced Cost and Experience, Random, or Round Robin.
  • Select tags, such as Text Generation.
  • Enter a description.

• Select Publish Immediately or Scheduled Publication.

• Select Next to open Step 2.

Step 2: Billing Configuration

• Select Free or Paid billing.
• For paid billing, select Uniform Billing, Input/Output Billing, or Tiered Billing.
• Configure prices in Credits per 1M tokens:
  • Original Input Price is the input reference price shown for comparison.
  • Input Sale Price is the actual settlement price for model input.
  • Original Output Price is the output reference price shown for comparison.
  • Output Sale Price is the actual settlement price for model output.

• Select Save Only or Submit for Review.

Parameter Reference - Aggregation Model

Field	Type	Example	Description
Model Type	Single select	Chat / Image	Required; aggregation model function
Model Subtype	Select	LLM	Required; specific model subtype
Member Models	List selection	Multiple provider instances	Required; select at least two published models
Enabled	Switch	On / Off	Required; controls whether the member receives traffic
Minimum Success Rate	Percentage	80%	Required; members below the threshold are excluded
Maximum Concurrency	Number	1500	Required; maximum concurrent requests for the member
Maximum Context Length	Number	128K	Required; member context limit
Input Token Cost	Number	2000	Required; cost per million input tokens
Output Token Cost	Number	8000	Required; cost per million output tokens
Custom Identifier	Text	Qwen3-235b-a22b-instruct-2507	Required; public aggregation identifier
Matching Policy	Single select	Cost First / Success Rate First / Balanced / Random / Round Robin	Required; request-routing policy
Tags	Select	Text Generation	Optional; aggregation tags
Description	Text	Aggregation model...	Optional; aggregation description
Publication Method	Single select	Immediate / Scheduled	Required; publication time
Billing Method	Single select	Free / Paid	Required; whether calls are billed
Billing Mode	Single select	Uniform / Input-Output / Tiered	Required for paid models
Original Input Price	Number	40.00 Credits/1M tokens	Optional; input reference price
Input Sale Price	Number	20.00 Credits/1M tokens	Required; actual input settlement price
Original Output Price	Number	160.00 Credits/1M tokens	Optional; output reference price
Output Sale Price	Number	80.00 Credits/1M tokens	Required; actual output settlement price

Completion Checklist

Purpose: These are the exit criteria for the current feature task. Use them to decide whether the result is observable and reviewable and whether you can continue to the next step in the scenario. They do not repeat the procedure; if any item fails, follow the troubleshooting section below.

Check	Pass Criteria
1	The aggregation contains at least two enabled member models.
2	The matching policy fits the business objective and member thresholds and costs are complete.
3	Publication or review status is correct.
4	A controlled request succeeds through the aggregation endpoint.
5	Remaining members accept traffic when one member is disabled.

Troubleshooting

Symptom	Check First
No member model can be selected	Publication state, modality, protocol, and visibility
Submission fails	Member count, required fields, billing, and publication area
Requests always select one member	Matching policy, enabled state, success-rate threshold, and cost
Aggregation calls fail	Member health, protocol compatibility, rate limits, and timeout

User Manual

• My Models
• My Models