Qwen3.5 9B Base
Ministral 14B
685B- Context
- 131Ktokens
- Released
- 1y ago
- Size
- 8Bparameters
DeepSeek R1 0528 Qwen3 8B is an open-weights, MIT-licensed model from DeepSeek, created by post-training Qwen3 8B Base on chain-of-thought data distilled from DeepSeek-R1-0528. DeepSeek reports a 10.0% improvement over Qwen3 8B on AIME 2024 and says it matches the performance of Qwen3-235B-thinking. The model targets math and reasoning tasks, with DeepSeek offering the distilled chain-of-thought for reasoning research and small-scale model development.
How it compares
Size class
Small8BAn open-weight model of 5 to 25 billion parameters, the size that runs on a single consumer GPU.
Open models of a similar size from other labs
Vs. your stack
Add the models you use to see how this one compares.
Sign inAdditional source evidence
- Source model
- deepseek/deepseek-r1-0528-qwen3-8b
- Source family
- deepseek-thinking
- Last updated
- 2025-05-29
- Model context
- 131K
- Model output limit
- 32K
- Open weights
- Yes
- File attachments
- No
- Reasoning
- Yes
- Tool calling
- No
- Temperature control
- Yes
Downloadable weights
Complete source record (JSON)
{
"id": "deepseek/deepseek-r1-0528-qwen3-8b",
"name": "DeepSeek R1 0528 Qwen3 8B",
"description": "DeepSeek R1 0528 distilled into Qwen3 8B for mathematics, coding, and reasoning",
"family": "deepseek-thinking",
"last_updated": "2025-05-29",
"attachment": false,
"reasoning": true,
"tool_call": false,
"temperature": true,
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"limit": {
"context": 131072,
"output": 32000
},
"release_date": "2025-05-29",
"open_weights": true,
"license": "MIT",
"weights": [
{
"label": "Hugging Face",
"url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B"
}
]
}Frequently asked questions
Running locally
Estimated 6.1 GB at Q4, 10.3 GB at Q8, or 18.9 GB at FP16. These figures come from the 8B parameter count and cover weights plus about 15% and 0.5 GB for a short context.
Yes. A 24 GB GPU fits at FP16 (estimated 18.9 GB), a 16 GB or 12 GB GPU fits at Q8 (estimated 10.3 GB), and an 8 GB GPU fits at Q4 (estimated 6.1 GB).
Yes. At Q4 the model needs an estimated 6.1 GB, and assuming macOS lets the GPU use about 75% of unified memory, the smallest common Mac memory size that fits is 16 GB.
Privacy and data
Yes. DeepSeek R1 0528 Qwen3 8B has open weights, so you can self-host it for full control of your data.
Capabilities and specs
The context window is 131K tokens (131,072), and the maximum output is 32K tokens (32,000).
No. DeepSeek R1 0528 Qwen3 8B only accepts text as input.
No. DeepSeek R1 0528 Qwen3 8B does not support tool calling (function calling).
Yes. DeepSeek R1 0528 Qwen3 8B supports a reasoning mode.
Quality and benchmarks
Magistral Small by Mistral AI, FastVLM 0.5B by Apple, and FastVLM 1.5B by Apple. They are matched on model kind and release date, with no comparable price.
Magistral SmallDeepSeek R1 0528 Qwen3 8B vs Magistral SmallFastVLM 0.5BDeepSeek R1 0528 Qwen3 8B vs FastVLM 0.5BFastVLM 1.5BDeepSeek R1 0528 Qwen3 8B vs FastVLM 1.5B
Versions and lineage
Both have a context window of 131K tokens (131,072). The difference is the release date: DeepSeek R1 0528 Qwen3 8B was released on May 29, 2025, while DeepSeek-R1-Distill-Qwen-32B was released on January 20, 2025.
DeepSeek-R1-Distill-Qwen-32BDeepSeek R1 0528 Qwen3 8B vs DeepSeek-R1-Distill-Qwen-32B
Licensing and access
It has open weights, released under the MIT license.
Yes. The MIT license is a permissive license that allows commercial use.
From the Hugging Face repository deepseek-ai/DeepSeek-R1-0528-Qwen3-8B.
Basics
DeepSeek made DeepSeek R1 0528 Qwen3 8B, and it was released on May 29, 2025.
Yes. DeepSeek R1 0528 Qwen3 8B has open weights, so you can download and run it yourself.
Generated from Vertical Tabs data, updated November 10, 2026.
Discussion
No comments yet
Be the first to share your thoughts.