DeepSeek R1 DistillOpen Weight

Released

DeepSeek R1 Distill is an AI model released by DeepSeek on Monday, Jan 20 2025, the same day as DeepSeek R1. It is an open-weight model — the trained weights are available to download and run. It was released in 1.5B and 7B and 8B and 14B and 32B and 70B parameter sizesThe learned values that shape how a model responds. B means billion and T means trillion. For mixture-of-experts models, the total count is larger than the number active for each token. Parameter count alone does not measure quality.. Benchmark results (shown below) cover LiveCodeBench, MATH-500, AIME 2024, and GPQA Diamond.

Benchmarks

Coding

LiveCodeBenchCompetitive coding — Coding problems published so recently the AI can't have seen them in training — a contamination-free test of raw programming skill. Higher is better.
57.5%
#37 of 60

Reasoning & science

MATH-500Reasoning — accuracy on the 500-problem MATH subset
94.5%
#7 of 15
AIME 2024Reasoning — single-answer accuracy on the 2024 invitational mathematics exam without tools
70%
#14 of 21
GPQA DiamondScience — Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better.
65.2%
#96 of 125

Compare DeepSeek R1 Distill with

DeepSeek R1 Distill

About

DeepSeek released six dense R1 distillations together on January 20, 2025. Training on examples generated by R1 transferred reasoning behavior into separately runnable models based on Qwen2.5 and Llama, ranging from 1.5B to 70B parameters.

This entry groups the simultaneous release across both foundation-model families. All displayed benchmark scores use the Llama 70B checkpoint, keeping the evaluation consistent instead of combining the best result from each size. The underlying Qwen and Llama license conditions also apply to their respective derivatives.

Frequently asked questions

DeepSeek R1 Distill was released by DeepSeek on Monday, Jan 20 2025.

All DeepSeek releases

30 tracked