Domestic large model 2025 Latest in December
GLM-4.7, MiniMax M2.1, DeepSeek, Qwen latest news
🔥 Latest release in December 2025
📅 December 23
GLM-4.7 Zhipu open source, Code Arena is the first open source
📅 December 24
MiniMax M2.1 Multi-language programming SOTA
🔷 GLM-4.7 Open Source No. 1 in China
Release time: December 23, 2025 | Company: Zhipu AI
Core Highlights
- Code Arena: First in open source, first in domestic production, surpassing GPT-5.2
- Context: 128K tokens
- Open source: Completely open source and can be deployed locally
- Skills module: Support unified planning of multi-modal tasks
💻 Coding ability
Enhanced coding scenarios, long-range mission planning
🔧 Tool collaboration
Multi-tool call and collaboration
💬 Conversation skills
Chatting, writing, role-playing improvements
📊 Low hallucination
Extremely low hallucination rate, high stability
Official website: open.bigmodel.cn
🔶 MiniMax M2.1 Open Source SOTA
Release time: December 24, 2025 | Company: MiniMax (Xiyu Technology)
Core Highlights
- Multi-SWE-bench: 49.4%, surpassing Claude Sonnet 4.5, global SOTA
- Parameters: 229B total parameters, 10B activation parameters (MoE)
- Multi-language programming: Rust/Java/Go/C++/TS/JS comprehensive improvement
- Open source: Hugging Face Open Source
🌐 Multi-language programming
Rust、Java、Go、C++、Kotlin、Objective-C、TS、JS
📱 Mobile Development
Enhanced native Android/iOS development capabilities
🎨Web Design
Web & App design understanding and aesthetic expression
⚡ Efficiency improvement
The thinking chain is simpler and Token consumption decreases
Official website: minimax.chat
🔵 DeepSeek Open Source
Latest version: V3-0324(2025.03.24)、R1(2025.01.20)
DeepSeek V3
685B parameters, MoE architecture, strong universal capabilities
DeepSeek R1
Inference model, thinking chain, mathematics/code expertise
Official website: deepseek.com (Free)
📊 Comparison of domestic large models (December 2025)
| Model | Release | Parameters | Features | Open Source |
|---|---|---|---|---|
| GLM-4.7 | 2025.12.23 | - | Code Arena No. 1 in open source | ✅ |
| MiniMax M2.1 | 2025.12.24 | 229B (10B activation) | Multi-language programming SOTA | ✅ |
| DeepSeek V3 | 2024.12.26 | 685B | Ultimate cost performance | ✅ |
| DeepSeek R1 | 2025.01.20 | - | Strong reasoning ability | ✅ |
| Qwen 2.5 | 2024.09 | 72B | Multimodal, long text | ✅ |
| Wenxin 4.5 | 2025 | - | Strong understanding of Chinese | ❌ |
💡 Usage suggestions
🆓 Free first choice
DeepSeek - The official website is free and the performance is close to GPT-4
💻 The first choice for programming
GLM-4.7 / M2.1 - Open source and locally deployable
🤖 Agent Development
GLM-4.7 - Skills module, tool collaboration
📱 Mobile Development
MiniMax M2.1 - iOS/Android native development
🔗 Quick access
- GLM: open.bigmodel.cn
- MiniMax: minimax.chat
- DeepSeek: deepseek.com (Free)
- Tongyi Qianwen: tongyi.aliyun.com
- A word from Wen Xin: yiyan.baidu.com