DeepSeek-R1-0528-Qwen3-8B

Name: DeepSeek-R1-0528-Qwen3-8B
Author: DeepSeek

DeepSeek · DeepSeek-R1 distill · released 2025-05-29 · data checked 2026-11-01 Needs verification

DeepSeek-R1-0528-Qwen3-8B is a 8.2B-parameter dense model from DeepSeek (China), released 2025-05-29 under the MIT license. It scores 53.1 on the Small LM Index, #9 of 26 small models. It needs about 6.8 GB of memory at 4-bit with an 8K-token context.

Distilled from DeepSeek-R1-0528 into Qwen3 8B Base.

Small LM Index

53.1

#9 of 26 scored small models

Likely 48.3–57.6 · from 3 of 6 core benchmarks

Parameters

8.2B

Dense

Price per 1M tokens

—

No hosted API on OpenRouter; self-host

Context window

128K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best small model on that benchmark. Sources: 3 reported by DeepSeek. How we label sources.

GPQA Diamond Lab-reported Knowledge61.1

Class best: Qwen3.5 9B 81.7

AIME Lab-reported Math · AIME 202576.3

Class best: Apriel 1.6 15B Thinker 88.0

LiveCodeBench Lab-reported Coding60.5

Class best: Granite 4.2 8B 73.2

Not reported: estimated

What the index model expects from this model's other scores. Not a measurement.

MMLU-Pro≈ 74

Humanity's Last Exam≈ 7

SWE-bench Verified≈ 35

Specifications

Lab: DeepSeek (CN · China)
Architecture: Dense transformer
Total parameters: 8.2B
Active parameters: All (dense)
Size class: Small
Context window: 128K tokens
Input modalities: text
Reasoning: Reasoning (thinking)
License: MIT
Openness: Open weights
Memory to run: ~6.8 GB at 4-bit, 8K context (more sizes)
Hugging Face: 357.7K downloads (30 days) · 1.1K likes

Weights on Hugging Face ↗

Compare with

DeepSeek-R1-0528-Qwen3-8B vs Qwen3.5 9B (9B, index 63.9)
DeepSeek-R1-0528-Qwen3-8B vs Qwen3 8B (8.2B, index 48.5)
DeepSeek-R1-0528-Qwen3-8B vs Granite 4.2 8B (8B, index 63.5)
DeepSeek-R1-0528-Qwen3-8B vs Ministral 3 8B (8B, index 55.2)
DeepSeek-R1-0528-Qwen3-8B vs Gemma 4 E4B (8B, index 38.2)
DeepSeek-R1-0528-Qwen3-8B vs MiniCPM-SALA (9.5B, index 52.6)

Sources

https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B

Something wrong here? Report it.

Run it locally

Uses unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF (Q4_K_M). Swap the tag for another size: BF16, IQ4_NL, IQ4_XS, Q2_K, Q2_K_L, Q3_K_M, Q3_K_S, Q4_0….

Ollama

ollama run hf.co/unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF:Q4_K_M

llama.cpp

llama-server -hf unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF:Q4_K_M

Then open http://localhost:8080 for a chat window.

MLX (Mac)

pip install -U mlx-lm
mlx_lm.chat --model mlx-community/DeepSeek-R1-0528-Qwen3-8B-4bit

vLLM (GPU server)

vllm serve deepseek-ai/DeepSeek-R1-0528-Qwen3-8B

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	32K context	128K context
16-bit F16	18 GB	22 GB	36 GB
8-bit Q8_0	10.0 GB	14 GB	28 GB
4-bit Q4_K_M	6.2 GB	10 GB	25 GB

Check it against your GPU, Mac or phone →