DeepSeek V3.1

deepseek-ai · released 2025-08-21 · mit license

DeepSeek V3.1 by deepseek-ai — 684.53B MoE (37B active) open-weight local LLM. License: mit. Benchmarks, VRAM requirements per quantization, GPU compatibility, pricing and community reviews on slopsome.com.

Key specs

TypeLocal open-weight
Parameters684.53B total · MoE, 37B active
Architecturedeepseek_v3
Context window164K tokens
Knowledge cutoff2025-06-01
Modalitiestext
Recommended backends
Minimum viable rig

Benchmark scores

GPQA Diamond74%
SWE-bench Verified66%
AIME88%
MMLU-Pro83%
BFCL v3 (tool use)75%
Composite score7.08
Community rating4.5★ (2 reviews, 0 net votes)

VRAM & disk per quantization

QuantVRAMDiskRAMContext
Q8700 GB690 GB768 GB131K
Q4_K_M398.5 GB397 GB448 GB164K

API pricing (per 1M tokens)

ProviderInputOutputFree tier
DeepSeek$0.27$1.1No
SiliconFlow$0.27$1.1No
Together AI$0.6$1.7No
Fireworks AI$0.9$0.9No