whatmodel.sh / DeepSeek R1 Distill Llama 8B

DeepSeek R1 Distill Llama 8B

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

DeepSeek R1 Distill Llama 8B is an open-weights language model from DeepSeek, released January 2025. Its published size is 8.03B parameters and a 131k-token context window. Licence: mit.

46.9±1.2#400 of 719 LLMsfrom 18 benchmark results across 7 sources

$ how it compares

The LLMs ranked either side of DeepSeek R1 Distill Llama 8B, out of 828 with a score, then more from DeepSeek.

#modelscorewhereRAMreleased
475Qwen1.5 110B Chat Alibaba47±1local~45GB2024-04
476WhiteRabbitNeo-33B-v1 WhiteRabbitNeo47±1local~22GB2023-12
477DeepSeek R1 Distill Llama 8B DeepSeek47±1local~6.2GB2025-01
478EXAONE 4.0 1.2B LG AI Research47±2local~2.2GB2025-07
479InternVL2-26B Shanghai AI Laboratory (OpenGVLab)47±2local~55GB2024-07
more from DeepSeek
21DeepSeek V4.1 Flash DeepSeek78±1api $0.26/M +local~537GB2026-09
31DeepSeek V4 Pro (0813) DeepSeek76±1api $1.63/M +local~1010GB2026-08
35DeepSeek V3.2 Speciale DeepSeek76±3comparison only

see all LLMs · find the ones that fit your machine

$ what it's good at

coding▁▁▁▃▆██▇▆▄▃▁ #429 of 458 · 22% of its score · 5 tests

Beats Qwen3.5 2B here, though they rank higher overall. Trails Llama 3.2 1B Instruct, SmolLM2 1.7B Instruct, Mistral 7B Instruct v0.1 here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
LiveCodeBench (version unspecified) lab39.6% ✎▄····▄▄··▄█▄ beats 2/6medium · avg 41%▇▇▇▇▇ 12%at its level
CodeReviewQA: change localisation (hard), invariant accuracy0.6% ✎█▂▁▁▂▂▂▃▅▅▅▇ beats 7/47medium · avg 36%▇▇ 3.8%sharp · at its level · noisy
CodeReviewQA: solution identification (hard), invariant accuracy6.8% ✎▃··▁▃▂▃▂█▃▄▂ beats 6/47medium · avg 54%▇▇ 3.8%sharp · at its level · noisy
CodeReviewQA: automated code refinement, exact match rate4.9% ✎▁▁▃▁▁▁▂▂▃█▅▄ beats 3/47hard · avg 33%▇ 2.7%at its level
CodeReviewQA: change type recognition, invariant accuracy0.4% ✎▃▁▁······▁▃█ beats 0/47easy · avg 67%▇ 0.1%blunt · at its level · noisy
math▁▂▃▅▆▅▄█▄▁▁▁ #204 of 398 · 31% of its score · 2 tests

Beats DeepSeek V3 (Dec '24), GPT-4o (2024-08-06), Command A (03-2025) here, though they rank higher overall. Trails Olmo 3 7B Instruct here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
AIME 2024 lab +250.4% ✎▃·▃·▃▃···█·▃ beats 2/7medium · avg 53%▇▇▇▇▇▇▇▇ 19%sharp · at its level · several runs
MATH-500 lab +289.1% ✎▄·▄▄·▄▄█···▄ beats 1/7very easy · avg 91%▇▇▇▇▇ 12%several runs
other tests5 not in a task · 46% of its score
testresultvs the others in its rundifficultycountswhy
GPQA Diamond lab +249.0% ✎▄·····█·█▄▄▄ beats 1/7hard · avg 47%▇▇▇▇▇▇▇▇▇▇▇▇▇ 33%sharp · several runs
BRIDGE clinical text, zero-shot average (87 tasks)28.5% ✎▁▁▁▁▁▂▂▃▄▆█▃ beats 24/97hard · avg 32%▇▇▇ 6.4%blunt · at its level
Open PT-LLM Leaderboard average (9 Portuguese tasks)42.4% ✎▃▁▁▁▁▁▃▁▂█▅▁ beats 50/276medium · avg 73%▇ 3.0%at its level · noisy
EuroEval HellaSwag (English), test split, 5-shot generative MC, accuracy53.7% ✎▃·▁▁▂▁▂▅█▇▄▁ beats 28/100medium · avg 68%▇ 2.4%at its level · noisy
EuroEval Life in the UK (English knowledge MC), test split, 5-shot, accuracy78.3% ✎▁▁▁▁▁▁▁▂▄█▄▃ beats 61/146easy · avg 83%▇ 1.7%at its level · noisy

$ where to run it

llama.cpp · recommended

version

  1. Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.
    macOS / Linux
    curl -LsSf https://llama.app/install.sh | sh
    Windows
    irm https://llama.app/install.ps1 | iex

    other ways to install ↗

  2. Run DeepSeek R1 Distill Llama 8B · needs ~4.9 GB · from huggingface.co/lmstudio-community
    chat in the terminal
    llama cli -hf lmstudio-community/DeepSeek-R1-Distill-Llama-8B-GGUF:Q4_K_M
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf lmstudio-community/DeepSeek-R1-Distill-Llama-8B-GGUF:Q4_K_M

LM Studio

version

  1. Install LM Studio · Open the app once before using the lms command.

    download it (macOS / Windows / Linux) ↗

  2. Run DeepSeek R1 Distill Llama 8B · needs ~4.9 GB · from huggingface.co/lmstudio-community

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/lmstudio-community/DeepSeek-R1-Distill-Llama-8B-GGUF@Q4_K_M -y
  4. Then run lms ls to see its name, and lms chat followed by that name

Ollama

version

  1. Install Ollama

    download it (macOS / Windows) ↗

    Linux
    curl -fsSL https://ollama.com/install.sh | sh
  2. Run DeepSeek R1 Distill Llama 8B · needs ~5.32 GB · from ollama.com
    chat in the terminal
    ollama run deepseek-r1:8b-llama-distill-q4_K_M

MLX (Mac)

version

  1. Install MLX (Mac) · Apple silicon Macs only, text only.
    with uv
    uv tool install mlx-lm
    or
    pip install mlx-lm
  2. Run DeepSeek R1 Distill Llama 8B · needs ~4.5 GB, Mac only · from huggingface.co/mlx-community
    chat in the terminal
    mlx_lm.chat --model mlx-community/DeepSeek-R1-Distill-Llama-8B-4bit
  3. Or start an OpenAI-compatible server at http://localhost:8080/v1
    mlx_lm.server --model mlx-community/DeepSeek-R1-Distill-Llama-8B-4bit

Weights and documentation: huggingface.co/deepseek-ai ↗

$ how the score was worked out · deepseek-ai/DeepSeek-R1 model… 34%, Microsoft — Phi-4-Mini Techni… 20%, microsoft/Phi-4-mini-reasonin… 20% and 4 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
deepseek-ai/DeepSeek-R1 model card · lab-reported4▇▇▇▇▇▇▇▇▇▇ 34%mit
Microsoft — Phi-4-Mini Technical Report3▇▇▇▇▇▇ 20%CC BY 4.0
microsoft/Phi-4-mini-reasoning model card3▇▇▇▇▇▇ 20%MIT
CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)4▇▇▇ 10%CC-BY-4.0
BRIDGE Medical Leaderboard (Harvard Medical School / Brigham and Women's YLab)1▇▇ 6%MIT
EuroEval leaderboards (results collection), English generative leaderboard2▇ 4%MIT
Open Portuguese LLM Leaderboard (archived)1▇ 3%Apache-2.0
prior · pull toward the middle▇ 1%
weighted average= 46.9
$ quantization evidence · 26 measurements

How much of the full-precision result a quantized version kept, measured within one study. Rows from the same study share a table and aren't independent.

precisiontestfull → quantizedkept
AWQ 4-bit (AutoAWQ; vLLM)AIME 2024 accuracy (mean of 3 passes; vLLM, T=0.6, top-p 0.95, 32768 max tokens)0.422 → 0.478113.3%source ↗
AWQ 4-bit (AutoAWQ; vLLM)Average of AIME 2024, FOLIO and Temporal accuracy (published 'Avg'; 3 passes each; vLLM, T=0.6, top-p 0.95, 32768 max tokens)0.652 → 0.666102.1%source ↗
FP8 W8A8 dynamic (FP8 weights symmetric per-channel, FP8 activations dynamic per-token; llm-compressor FP8_DYNAMIC)AIME 2024 pass@10.492 → 0.508103.2%source ↗
FP8 W8A8 dynamic (FP8 weights symmetric per-channel, FP8 activations dynamic per-token; llm-compressor FP8_DYNAMIC)IFEval 0-shot instruction-level strict acc (OpenLLM v2 config, chat template)0.384 → 0.387100.8%source ↗
FP8 W8A8 dynamic (FP8 weights symmetric per-channel, FP8 activations dynamic per-token; llm-compressor FP8_DYNAMIC)Reasoning average of 3 tasks (AIME 2024 pass@1, MATH-500 pass@1, GPQA Diamond pass@1)0.629 → 0.633100.6%source ↗
FP8 W8A8 dynamic (FP8 weights symmetric per-channel, FP8 activations dynamic per-token; llm-compressor FP8_DYNAMIC)MATH-500 pass@10.902 → 0.902100.1%source ↗
FP8 W8A8 dynamic (FP8 weights symmetric per-channel, FP8 activations dynamic per-token; llm-compressor FP8_DYNAMIC)GPQA Diamond pass@10.493 → 0.48798.9%source ↗
GPTQ 3-bit (vLLM)Average of AIME 2024, FOLIO and Temporal accuracy (published 'Avg'; 3 passes each; vLLM, T=0.6, top-p 0.95, 32768 max tokens)0.652 → 0.47873.3%source ↗
GPTQ 3-bit (vLLM)AIME 2024 accuracy (mean of 3 passes; vLLM, T=0.6, top-p 0.95, 32768 max tokens)0.422 → 0.11126.3%source ↗
GPTQ 4-bit (vLLM)AIME 2024 accuracy (mean of 3 passes; vLLM, T=0.6, top-p 0.95, 32768 max tokens)0.422 → 0.422100.0%source ↗
GPTQ 4-bit (vLLM)Average of AIME 2024, FOLIO and Temporal accuracy (published 'Avg'; 3 passes each; vLLM, T=0.6, top-p 0.95, 32768 max tokens)0.652 → 0.58189.1%source ↗
INT4 W4A16 (weights only, symmetric per-group, group size 128; card states GPTQ via llm-compressor)GSM8K 5-shot strict-match (lm-evaluation-harness, vLLM)0.628 → 0.631100.5%source ↗
INT4 W4A16 (weights only, symmetric per-group, group size 128; card states GPTQ via llm-compressor)MATH-500 pass@10.902 → 0.89999.7%source ↗
INT4 W4A16 (weights only, symmetric per-group, group size 128; card states GPTQ via llm-compressor)IFEval 0-shot instruction-level strict acc (OpenLLM v2 config, chat template)0.384 → 0.37998.6%source ↗
INT4 W4A16 (weights only, symmetric per-group, group size 128; card states GPTQ via llm-compressor)OpenLLM v1 average of 6 tasks (ARC-Challenge 25-shot, GSM8K 5-shot, HellaSwag 10-shot, MMLU 5-shot, TruthfulQA MC2 0-shot, Winogrande 5-shot; lm-evaluation-harness `openllm`, vLLM)0.599 → 0.58397.3%source ↗
INT4 W4A16 (weights only, symmetric per-group, group size 128; card states GPTQ via llm-compressor)Reasoning average of 3 tasks (AIME 2024 pass@1, MATH-500 pass@1, GPQA Diamond pass@1)0.629 → 0.61197.2%source ↗
INT4 W4A16 (weights only, symmetric per-group, group size 128; card states GPTQ via llm-compressor)GPQA Diamond pass@10.493 → 0.47195.6%source ↗
INT4 W4A16 (weights only, symmetric per-group, group size 128; card states GPTQ via llm-compressor)AIME 2024 pass@10.492 → 0.46394.0%source ↗
INT4 W4A16 (weights only, symmetric per-group, group size 128; card states GPTQ via llm-compressor)MMLU 5-shot acc (lm-evaluation-harness, vLLM)0.556 → 0.52193.6%source ↗
INT8 W8A8 (weights symmetric per-channel, activations symmetric per-token; SmoothQuant + GPTQ via llm-compressor)IFEval 0-shot instruction-level strict acc (OpenLLM v2 config, chat template)0.384 → 0.384100.1%source ↗
INT8 W8A8 (weights symmetric per-channel, activations symmetric per-token; SmoothQuant + GPTQ via llm-compressor)MATH-500 pass@10.902 → 0.999.8%source ↗
INT8 W8A8 (weights symmetric per-channel, activations symmetric per-token; SmoothQuant + GPTQ via llm-compressor)MMLU 5-shot acc (lm-evaluation-harness, vLLM)0.556 → 0.55599.8%source ↗
INT8 W8A8 (weights symmetric per-channel, activations symmetric per-token; SmoothQuant + GPTQ via llm-compressor)AIME 2024 pass@10.492 → 0.49199.6%source ↗
INT8 W8A8 (weights symmetric per-channel, activations symmetric per-token; SmoothQuant + GPTQ via llm-compressor)Reasoning average of 3 tasks (AIME 2024 pass@1, MATH-500 pass@1, GPQA Diamond pass@1)0.629 → 0.62799.6%source ↗
INT8 W8A8 (weights symmetric per-channel, activations symmetric per-token; SmoothQuant + GPTQ via llm-compressor)OpenLLM v1 average of 6 tasks (ARC-Challenge 25-shot, GSM8K 5-shot, HellaSwag 10-shot, MMLU 5-shot, TruthfulQA MC2 0-shot, Winogrande 5-shot; lm-evaluation-harness `openllm`, vLLM)0.599 → 0.59599.3%source ↗
INT8 W8A8 (weights symmetric per-channel, activations symmetric per-token; SmoothQuant + GPTQ via llm-compressor)GSM8K 5-shot strict-match (lm-evaluation-harness, vLLM)0.628 → 0.62198.9%source ↗

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json