DeepSeek-V3.1
Open Weights
DeepSeek
DeepSeek's August 2025 hybrid model, a 685B-parameter mixture-of-experts (37B active) that unifies chat, reasoning, and coding in a single checkpoint with switchable thinking. Open-weight and inexpensive to serve, it offers a 128K context window at a fraction of proprietary frontier pricing.
Strengths
- Unifies chat, reasoning, and coding with switchable thinking
- 685B-parameter MoE with only 37B active per token
- Open weights under permissive MIT license
- 128K context at a fraction of proprietary frontier pricing
- Strong multilingual and long-context handling
Caveats
- Text-only (no vision or audio input)
- Full-precision self-hosting demands substantial GPU memory
- Superseded by DeepSeek-V3.2 for long-context efficiency
- Vendor
- DeepSeek
- Context Window
- 128,000 tokens
- License
- MIT
- Release Date
- 2025-08-21
- Modalities
- text
Capabilities
Vision
Audio
Video
Tool Use
Resources
Pricing
$0.56 per 1M input tokens, $1.68 per 1M output tokens (non-reasoning) via the DeepSeek API. Open weights under MIT - free to self-host.