DeepSeek R1 0528

deepseek-ai · released 2025-05-28 · mit license

DeepSeek-R1-0528 (May 2025) is a reasoning-focused open-weights MoE that pushed AIME 2025 to 87.5% and GPQA Diamond to 81.0% per its model card, a top open-weights lab at the time.

Key specs

TypeLocal open-weight
Parameters684.53B total · MoE, — active
Architecturedeepseek_v3
Context window164K tokens
Knowledge cutoff2025-03-31
Modalitiestext
Recommended backends
Minimum viable rig

Benchmark scores

GPQA Diamond81%
SWE-bench Verified57.6%
AIME87.5%
MMLU-Pro85%
BFCL v3 (tool use)
Composite score6.42
Community ratingNo reviews yet

VRAM & disk per quantization

QuantVRAMDiskRAMContext
Q4_K_M398.5 GB397 GB164K

Strengths & weaknesses

Strengths: Top-tier open-weights reasoning at release (AIME 87.5, GPQA 81.0); Big jump in deep reasoning vs original R1; MIT, distillation-friendly

Weaknesses: High token usage / latency from long chain-of-thought; Surpassed by V3.1/V3.2 on agentic coding