Back to Models
DeepSeekOpen Source
DeepSeek-V3.1
DeepSeek-V3.1, developed by DeepSeek-AI, is a large-scale foundation model with an approximate parameter scale of 6.71 trillion. It features a long 128K context window and provides advanced chat capabilities.
Parameters
6710.0B
Context Window
128K
License
MIT
Release Date
2025-08-20
API Pricing
API pricing for this model is not yet available
Strengths
- ・Extremely large parameters of 6710B
- ・Long 128K context window
- ・High openness under MIT license
Weaknesses
- ・Massive model size of 1340GB
- ・Requires very advanced computational resources to run
- ・Difficult to operate on lightweight devices
Use Cases
- ・Building advanced conversational AI chatbots
- ・Long document analysis and processing
- ・Customization of large-scale foundation models
Deep Analysis
Parameters
671B MoE
GPQA Diamond
80.1% (thinking)
AIME 2025
88.4% (thinking)
SWE-Bench Verified
66.0%
Pricing
$0.28/$0.42 per 1M tokens
Release Date
August 21, 2025
Strengths
- ・Hybrid thinking/non-thinking mode
- ・Strong agent and tool-use
- ・Comparable to R1-0528 with faster responses
- ・Extremely competitive pricing
- ・MIT license
Weaknesses
- ・Safety alignment concerns (Azure)
- ・Non-thinking mode significantly weaker
- ・SWE-bench trails frontier models
- ・Superseded by V3.2
Competitor Comparison
| Model | SWE | GPQA | Price |
|---|---|---|---|
| Claude Sonnet 4.5 | 77.2% | 83.4% | $3/$15 |
| GPT-5 | 74.9% | 85.7% | $2.50/$15 |
| DeepSeek-V3.2 | 73.1% | 82.4% | $0.28/$0.42 |
DeepSeek-V3.1 is a hybrid thinking/non-thinking MoE model. 671B params, strong agent capabilities, extremely competitive pricing.
Sources
Analysis generated: 2026-05-24