Model Overview
DeepSeek-R1-Distill-Qwen-32B is a dense reasoning model with approximately 32 billion parameters released by DeepSeek. It is based on Qwen2.5-32B and fine-tuned using samples generated by DeepSeek-R1. The model weights have been officially released, and the model card lists the license as MIT. (github.com)
Capabilities and Uses
The model is intended for text generation and reasoning and can be used for solving mathematical problems, code generation and analysis, question answering, and research experiments. Official evaluations report results on mathematics, science question-answering, and programming benchmarks; actual performance depends on the task and deployment configuration. The weights can be used for local deployment, and the official repository provides launch examples for vLLM and SGLang. (github.com)
Limitations
The model may produce incorrect or unreliable reasoning and answers; important results should be verified by a human. Local deployment requires computing resources appropriate to the model's size; reasoning length and operating costs depend on the deployment configuration. The official recommendation is to use a temperature of 0.5–0.7, with 0.6 recommended, to reduce the risk of repetition or incoherent output. (github.com)
Date
Record date: 2025-01-20, based on the model repository creation date. Official metadata on the model hosting platform records that the repository was created on 2025-01-20 at 09:19 UTC; DeepSeek's announcement from the same period states that it released six distilled models alongside R1, and the official repository lists this 32B version. (huggingface.co)
Documentation note: In the current data, the exact standalone initial release date of this specific version has not been independently verified. The date above is based on the model repository creation time and is not asserted to be an initial release date earlier than all other channels. (huggingface.co)
Pricing
As of 2026-11-20, DeepSeek's official domestic RMB and international USD pricing pages have been checked, but no official API input or output prices for this specific distilled version have been verified on either page; prices for other DeepSeek models are not applied to this model. Hardware and hosting costs for local operation are separate. (api-docs.deepseek.com)