DeepSeek: DeepSeek V3.1
Model Type
Proprietary Model
API access only
Recommended Use Cases
Text Generation
Try DeepSeek V3.1
A hybrid reasoning model supporting both thinking and non-thinking modes within a single model (August 2025). V3.1 significantly improves tool calling and agent tasks while achieving comparable reasoning quality to DeepSeek-R1 with faster response times.
Per DeepSeek:
DeepSeek-V3.1 is a hybrid model that supports both thinking mode and non-thinking mode. DeepSeek-V3.1-Think achieves comparable answer quality to DeepSeek-R1-0528, while responding more quickly.
Role in V3.1 Series
V3.1 is the main instruction-tuned model of the series, balancing reasoning capabilities with practical efficiency. It introduces hybrid thinking modes and significantly improves agentic capabilities over V3.
Key Features
- Architecture: 685B MoE (671B parameters, 37B active)
- Context Window: 128K tokens
- Hybrid Modes: Toggle between thinking and non-thinking via chat template
- License: MIT
Key Improvements
- Hybrid Thinking Mode: Single model supports both modes by changing the chat template
- Smarter Tool Calling: Optimized performance in tool usage and agent tasks
- Higher Thinking Efficiency: Comparable to R1-0528 quality with faster responses
Agent Capabilities
- Search tool via multi-turn tool-calling process
- Code agent support
- Structured tool calling