DeepSeek-R1-0528 you can run on your computer

2025-05-29

TLDR: You can run the small version of the latest DeepSeek R1 model on your computer with as little as ~6GB of RAM. Download the model from here. You need LM Studio 0.3.16 or newer.

undefined

DeepSeek-R1-0528-Qwen3-8B, an 8B model distilled from CoT of the new DeepSeek R1

New DeepSeek R1: DeepSeek-R1-0528

Yesterday (May 28, 2025) DeepSeek released a new iteration of their R1 model: DeepSeek-R1-0528. By leveraging more compute and improved post-training algorithms, this version greatly expands the model's reasoning and inference capabilities.

It scores highly on benchmarks ranging from math to programming and general logic, approaching results from models like O3 and Gemini 2.5 Pro.

undefined

DeepSeek-R1-0528 benchmarks from the release announcement

Distilled 8B Model: DeepSeek-R1-0528-Qwen3-8B

As a part of R1-0528 release, DeepSeek also shared DeepSeek-R1-0528-Qwen3-8B, a "distilled" model created by continuing the post-training of Qwen3 8B Base using chain-of-thought traces from DeepSeek-R1-0528. This smaller model also supports both tool use and reasoning.

The DeepSeek team reports that it achieves state-of-the-art results on AIME 2024, outperforming Qwen3 8B by about 10% and matching the much larger Qwen3-235B-thinking.

AIME 24AIME 25HMMT Feb 25GPQA DiamondLiveCodeBench (2408-2505)
Qwen3-235B-A22B85.781.562.571.166.5
Qwen3-32B81.472.9-68.4-
Qwen3-8B76.067.3-62.0-
Phi-4-Reasoning-Plus-14B81.378.053.669.3-
Gemini-2.5-Flash-Thinking-052082.372.064.282.862.3
o3-mini (medium)79.676.753.376.865.9
DeepSeek-R1-0528-Qwen3-8B86.076.361.561.160.5
Source: DeepSeek-R1-0528 on Hugging Face

You can run the distilled model with as little as 4GB of RAM. It supports tools and advanced reasoning out of the box.

This model is available in both MLX and GGUF formats and requires LM Studio 0.3.16 or newer. Open in LM Studio to download it on your Mac, Windows, or Linux machine.




👾 Join us

Interested in working on making local AI accessible and useful to everyone?

See open positions: https://lmstudio.ai/careers