Agent
1. Introduction of Sanplex features
1.1. Core Management Framework
1.1.1 Program
1.1.2 Project
1.1.3 Product
1.1.4 Execution
1.1.5 Management Models
1.2. Dashboard
1.2.1 Getting Started with Tutorials
1.2.2 Quick Add
1.2.3 Notification Center
1.2.4 Effort Tracking
1.2.5 Dashboard - Project and Execution
1.2.6 Dashboard - My Work, Contribution, and Recents
1.2.7 Dashboard - Approval
1.2.8 Contacts
1.3. Program
1.3.1 Program List
1.3.2 Program Kanban
1.3.3 Create Program
1.3.4 Create Sub-Programs
1.3.5 Project Charter Management
1.4. Product
1.4.1 Create Product
1.4.2 Manage Modules
1.4.3 Multi-Branch and Multi-Platform Management
1.4.4 Manage Plans
1.4.5 Manage Requirements
1.4.6 Requirement Reviews
1.4.7 Requirement Status and Phase
1.4.8 Create Releases
1.4.9 Track Progress
1.4.10 Business Requirements and Multi-Level Requirements
1.5. Project
1.5.1 Scrum Project
1.5.2. Waterfall Project
1.5.2.1 Manage Project Phases
1.5.2.2 Project Design
1.5.2.3 Project Matrix
1.5.2.4 Project Reports and Earned Value Management
1.5.2.5 Project Baselines
1.5.2.6 Project Change
1.5.3. Kanban Project
1.5.3.1 Set Up Kanban Projects
1.5.3.2 Configure Kanban Boards
1.5.3.3 Use Kanban Boards
1.5.4 Hybrid Agile Project
1.5.5 Hybrid Waterfall Project
1.5.6. Project Settings
1.5.6.1 Project Setup
1.5.6.2 Project Executions
1.5.6.3 Project Requirements
1.5.6.4 Project Builds
1.5.6.5 Project Tests
1.5.6.6 Project Docs
1.5.6.7 Project Releases
1.5.6.8 Project Reports
1.5.6.9 Project Risks
1.5.6.10 Project Issues
1.5.6.11 Project Opportunities
1.6. Admin Settings
1.7. Workflow

Models

2026-09-04 13:51:34
Sanplex Content
177
Last edited by Kelsea Zhang on 2026-09-04 14:06:45
Share links
Summary: This page covers how to use the model list, the models tested locally, and benchmark results and recommendations for embedding models used with the Knowledge Library.

Catalog

The models synchronize model information from the model list in the ZAI service console. Here, you can view all models integrated into ZAI services. Click the Converse button in the Actions column to begin a conversation with a model immediately.

Models page

I. Locally Tested Large Models: Categories and Vendors

1. Chat and General-Purpose Large Language Models (Chat/LLM)

Vendor / Organization Model Primary Use / Highlights
OpenAI gpt-5.2-chat High-quality general-purpose chat, reasoning, and writing
gpt-5-mini Lightweight general-purpose chat
gpt-4o-mini Multimodal, low latency
gpt-oss Open-source, ecosystem-oriented model
Zhipu AI GLM-4.5 General-purpose model
GLM-4.5-Air Lightweight and suited to high concurrency
GLM-4.5-X Enhanced-capability version
GLM-4.6 Next-generation flagship
CharGLM-4 Character- and persona-based conversation
Anthropic claude-4.5-opus Flagship model with strong reasoning
claude-4.5-sonnet Well-balanced performance
DeepSeek deepseek-chat General-purpose chat
DeepSeek-V3.2 Next-generation general-purpose model
deepseek-reasoner Enhanced reasoning
DeepSeek-R1 Reasoning-focused model
deepseek-r1_32b Large-parameter reasoning model
Alibaba (Qwen) qwen3 General-purpose base model
qwen-max High-performance version
qwen-turbo Budget-friendly option
qwen-flash Low latency
qwen-coder-plus Code generation
Moonshot kimi-k2-0905-preview Long-context processing and reasoning
kimi-k2-turbo-preview Low-latency version
Meta Llama-3.2 Open-source general-purpose LLM
Google gemma3 Lightweight general-purpose model
MiniMax MiniMax-M2 General-purpose conversation
Xiaomi mimo-v2-flash Fast responses
xiaomi-mimo-v2-flash Xiaomi ecosystem model
Mistral ministral-3 Lightweight and efficient
ByteDance doubao-seed-1.6 General-purpose model
doubao-seed-code Code-focused model

2. Vector / Embedding Models (Text Embeddings)

Vendor / Organization Model Primary Use / Highlights
OpenAI text-embedding-3-small General-purpose semantic embeddings
text-embedding-ada-002 Classic embedding model
Zhipu AI GLM-Embedding-2 Chinese and general-purpose embeddings
GLM-Embedding-3 Next-generation embedding model
Alibaba (Qwen) qwen3-embedding General-purpose embeddings
qwen-text-embedding-v1 Embedding v1
qwen-text-embedding-v2 Embedding v2
qwen-text-embedding-v3 Embedding v3
qwen-text-embedding-v4 Latest embedding version
BAAI (Beijing Academy of Artificial Intelligence) BAAI_bge-large-zh-v1.5 High-quality Chinese embeddings
bge-m3 Multilingual and multimodal
Snowflake snowflake-arctic-embed General-purpose embeddings
snowflake-arctic-embed2 Upgraded version
Jina AI jina-embeddings-v2-base-zh Chinese embeddings
Nomic nomic-embed-text Primarily for English
Mixedbread AI (MXBAI) mxbai-embed-large High-dimensional embeddings
NLP4All nlp_corom_sentence-embedding_chinese-base Chinese sentence embeddings
nlp_gte_sentence-embedding_chinese-base GTE Chinese base model
nlp_gte_sentence-embedding_chinese-large GTE Chinese large model
DeepSeek dmeta-embedding-zh Chinese semantic embeddings
ByteDance doubao-embedding General-purpose embeddings
doubao-embedding-large Higher-precision embeddings
BCE bce-embedding-base_v1 Chinese embeddings

3. Hybrid / Specialized Models

Model Vendor Positioning
bge-m3 BAAI Embeddings with multilingual / cross-modal support
embeddinggemma Google Embedding-first model with room for extension
qwen3 Alibaba (Qwen) Base model with multitask capabilities

II. Knowledge Library Embedding Benchmarks and Recommendations

These results come from a global embedding benchmark on 589 Sanplex objects in an internal test environment, including requirements, bug tickets, and technical manuals. The findings are intended to guide embedding-model selection for Chinese Knowledge Bases, with detailed scores provided below.

1. Recommended Models

Scenario Model (with size) Why it is recommended Score
Local deployment (best performance) qwen3-embedding (8b) Strongest local option; it matches Alibaba Cloud's hosted model and clearly outperforms the other small local models. 58
Local deployment (limited resources) bge-m3 (567m) Lightweight first choice. It scores highest when GPU memory is limited and is budget-friendly to run. 37
Cloud API (best possible quality) qwen-text-embedding-v3 Overall winner. Alibaba Cloud's model handles Chinese terms such as "Xuanxuan" and "ZAI" most accurately, narrowly edging out OpenAI. 72

2. Overall Ranking

The table below shows the final scores from the hands-on benchmark:

Rank Model Deployment Model Size Total Score Summary
1 qwen-text-embedding-v3 Alibaba Cloud API - 72 Best Chinese-language performance
2 text-embedding-ada-002 OpenAI API - 71 Very stable; just one point behind the leader
3 qwen-text-embedding-v4 Alibaba Cloud API - 58 Newer version, but slightly behind v3
3 qwen3-embedding Local (Ollama) 8b 58 Exceptional local result, matching the cloud model
5 dou-embedding Volcano Engine API - 45 Solid but unexceptional performance
6 doubao-embedding-large Volcano Engine API - 40 The larger version scored lower on this dataset
7 q-text-embedding-v1 Alibaba Cloud API - 37 Older version, gradually being phased out
7 bge-m3 Local (Ollama) 567m 37 Small-model standout and the first choice when resources are tight
9 mxbai-embed-large Local (Ollama) 335m 36 Just behind bge-m3
10 text-embedding-3-small OpenAI API - 34 OpenAI's lightweight option, with somewhat weaker Chinese-language performance
Write a Comment
Comment will be posted after it is reviewed.