Back to Models
DeepSeekOpen Source
DeepSeek-V3.1 Terminus
DeepSeek-V3.1 Terminus, developed by DeepSeek-AI, is a large foundation model boasting approximately 6.7 trillion parameters. It features a context window of 128K and provides advanced chat capabilities.
Parameters
6710.0B
Context Window
128K
License
MIT
Release Date
2025-09-22
API Pricing
API pricing for this model is not yet available
Strengths
- ・Overwhelming parameter scale
- ・Long context window
- ・Open MIT license
Weaknesses
- ・Very large model size
- ・Requires extensive storage capacity
- ・High computational resource demand
Use Cases
- ・Development of advanced conversational AI
- ・Analysis of large-scale data
- ・Understanding and processing complex context
Deep Analysis
Parameters
684.5B MoE
GPQA Diamond
74.9%
SWE-Bench Verified
66.0%
MMLU-Redux
91.8%
Pricing
$0.30/$0.95 per 1M tokens
Release Date
September 22, 2025
Strengths
- ・Strong coding agent (66% SWE-Bench)
- ・Excellent general knowledge (91.8% MMLU)
- ・Extended long-context training
- ・MIT license
Weaknesses
- ・Non-thinking mode only
- ・AIME 2025 modest (49.8%)
- ・Trails frontier by 10+ pp
- ・Quickly superseded
Competitor Comparison
| Model | GPQA | Price |
|---|---|---|
| V3.1 | 80.1% | $0.28/$0.42 |
| V3.2 | 82.4% | $0.28/$0.42 |
DeepSeek-V3.1 Terminus is an extended pre-training variant with 684.5B params trained on 6.7T tokens.
Sources
Analysis generated: 2026-05-24