Model
Model explorer

DeepSeek-R1-0528

OPEN
DeepSeek · DeepSeek-R1 family · released May 28, 2025

Major R1 refresh with deeper chain-of-thought (AIME24 79.8 to 91.4), fewer hallucinations, and added JSON/function calling; approached o3/Gemini 2.5 Pro.

ReasoningCodingVisionFunction callingTool useAgentic
1757.4
Elo · rank #83
Parameters
671B
Active params
37B (MoE)
Context
128K tokens
Architecture
DeepSeekMoE: 256 routed + 1 shared expert, top-8 routing, 61 layers (3 dense + 58 MoE), Multi-head Latent Attention (MLA)
License
MIT
Languages
API price (in/out)
$0.55 / $2.19
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
Aider PolyglotCoding71.6%#13
best: Claude Opus 4.5 · 89.4%
AIMEMath91.4%#41
best: GPT-5.2 · 100.0%
ARC-AGI-1Reasoning21.2%#15
best: GPT-5.6 · 96.5%
ARC-AGI-2Reasoning1.1%#27
best: GPT-5.6 · 92.5%
BFCL v3Agents37.0%#31
best: Hunyuan-A13B · 78.3%
CodeforcesCoding1930#24
best: DeepSeek-V4-Pro (Think Max) · 3206
GPQA DiamondReasoning81.0%#73
best: GPT-5.6 · 94.6%
Humanity's Last ExamReasoning17.7%#67
best: Claude Sonnet 5 · 57.4%
LiveCodeBenchCoding73.3%#60
best: DeepSeek-V4-Pro (Think Max) · 93.5%
MMLU-ProKnowledge85.0%#26
best: Claude Fable 5 · 91.5%
MMLU-ReduxKnowledge93.4%#12
best: Qwen3.7-Max · 95.0%
SimpleQAKnowledge27.8%#16
best: GPT-4.5 · 62.5%
SWE-bench VerifiedCoding57.6%#81
best: Claude Fable 5 · 95.0%
tau-benchAgents53.5%#19
best: Claude Opus 4.1 · 82.4%
Run it locally
VRAM @ Q4
404 GB
VRAM @ FP16
1342 GB
Fits on (Q4)
Multi-node cluster required
671B MoE far exceeds a single 24GB 4090; Unsloth's dynamic quants shrink it to ~185GB for CPU-offload rigs, still requiring far more than one consumer GPU.
Quantizations
GGUF Q4_K_M · GGUF IQ1/IQ2 (Unsloth Dynamic) · AWQ · MLX
API price $0.55/$2.19 · each benchmark row carries its own source badge (see methodology)