DeepSeek-R1-Distill-Qwen-32B

DeepSeek-R1-Distill-32B is a distilled model built upon Qwen2.5-32B, incorporating the best reasoning algorithms from DeepSeek-R1 and expert knowledge. It sets new records among open-source dense models across several reasoning benchmarks: AIME 2024–72.6%, MATH-500–94.3%, and others. In practice, this model is nearly on par with the distilled 70-billion-parameter version and even surpasses it in certain tests.

Technically, the model is designed for solving expert-level tasks: complex mathematical computations, code generation and analysis, scientific research, and processing long or intricate contexts. DeepSeek-R1-Distill-32B can be integrated into enterprise systems, cloud services, and platforms for automating intellectual labor. For end-user applications, it is indispensable for building expert systems, scientific assistants, platforms for automating complex business processes, and educational solutions that require thorough and well-articulated explanations in responses.

DeepSeek-R1-Distill-32B is the choice for those seeking maximum performance among open-source models without the need to move to the heaviest systems.

Announce Date: 20.01.2025
Parameters: 32.8B
Context: 131K
Attention Type: Full or Sliding Window Attention
VRAM requirements: 47.3 GB using 4 bits quantization
Developer: DeepSeek
Transformers Version: 4.43.1
License: Apache 2.0

Use our pre-built public endpoints to test inference and explore DeepSeek-R1-Distill-Qwen-32B capabilities.

Rent your own physically dedicated instance with hourly or long-term monthly billing.

We recommend deploying private instances in the following scenarios:

• maximize endpoint performance,
• enable full context for long sequences,
• ensure top-tier security for data processing in an isolated, dedicated environment,
• use custom weights, such as fine-tuned models or LoRA adapters.

Contact our dedicated neural networks support team at nn@immers.cloud or send your request to the sales department at sale@immers.cloud.