whatmodel.sh / DeepSeek R1 Distill Qwen 1.5B

DeepSeek R1 Distill Qwen 1.5B

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

DeepSeek R1 Distill Qwen 1.5B is an open-weights language model from DeepSeek, released January 2025. Its published size is 1.78B parameters and a 131k-token context window. Licence: mit.

37.2±1.7#574 of 719 LLMsfrom 16 benchmark results across 7 sources

$ how it compares

The LLMs ranked either side of DeepSeek R1 Distill Qwen 1.5B, out of 828 with a score, then more from DeepSeek.

#modelscorewhereRAMreleased
665Yi-1.5 6B (base) 01.AI37±3local~5.1GB2024-05
666Qwen2-VL 2B Instruct Alibaba37±1local~2.8GB2024-08
667DeepSeek R1 Distill Qwen 1.5B DeepSeek37±2local~2.6GB2025-01
668Gemma 2 2B IT Google37±1local~3.1GB2024-07
669Qwen2.5 Coder 0.5B Instruct Alibaba37±1local~1.8GB2024-11
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▁▃▁▃▆██▇▆▄▃▁ #455 of 458 · 26% of its score · 5 tests

Trails GPT-Neo 2.7B, Phi-1 (1.3B), OpenAssistant SFT-4 Pythia 12B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
LiveCodeBench (version unspecified) lab16.9% ✎▄····▄▄··▄█▄ beats 0/6medium · avg 41%▇▇▇▇▇▇ 15%typical
CodeReviewQA: solution identification (hard), invariant accuracy0.0% ✎▃··▁▂▂▃▂█▃▄▂ beats 2/47medium · avg 54%▇▇ 4.3%sharp · noisy
CodeReviewQA: automated code refinement, exact match rate2.2% ✎▃▁▁▁▁▁▂▂▃█▅▄ beats 0/47hard · avg 33%▇ 3.6%typical
CodeReviewQA: change localisation (hard), invariant accuracy0.0% ✎█▂▁▁▂▂▂▃▅▅▅▇ beats 3/47medium · avg 36%▇ 3.0%sharp · noisy
CodeReviewQA: change type recognition, invariant accuracy0.0% ✎▃▁▁······▁▃█ beats 1/47easy · avg 67%▇ 0.1%blunt · at its level · noisy
math▁▂▃▅▆▅▄█▄▁▁▁ #238 of 398 · 34% of its score · 4 tests

Beats Command A (03-2025), GPT-4o (2024-11-20), GPT-4o (2024-05-13) here, though they rank higher overall.

testresultvs the others in its rundifficultycountswhy
MATH-500 lab83.9% ✎▄·▄▄·▄▄█···▄ beats 0/7very easy · avg 91%▇▇▇▇▇▇▇▇ 20%typical
THINK-Bench final-answer accuracy (reasoning models)62.9% ✎▄█·▄····▄▄·█ beats 0/7easy · avg 76%▇▇▇ 8.2%at its level
AIME 2024 lab28.9% ✎▃·▃·▃▃···█·▃ beats 0/7medium · avg 50%▇▇ 5.7%sharp · noisy
OTIS mock AIME 2024–202521.4% ✎▃▃▄▃▃▃▇▇█▃▂▅ beats 59/193very hard · avg 8.5%▇ 0.6%sharp · too hard for it · noisy
other tests6 not in a task · 38% of its score
testresultvs the others in its rundifficultycountswhy
GPQA Diamond +133.6% ✎▁▁▁▁▃▄▆▄▅▇█▅ beats 24/208hard · avg 47%▇▇▇▇▇▇ 16%sharp · too hard for it · several runs
BRIDGE clinical text, zero-shot average (87 tasks)14.3% ✎▁▁▃▁▁▂▂▃▄▆█▃ beats 3/97hard · avg 32%▇▇▇▇ 10%blunt
Open PT-LLM Leaderboard average (9 Portuguese tasks)25.3% ✎▃▁▁▁▁▁▁▁▂█▅▁ beats 26/276medium · avg 73%▇▇ 5.1%at its level · noisy
EuroEval Life in the UK (English knowledge MC), test split, 5-shot, accuracy29.9% ✎▁▁▁▃▁▁▁▂▄█▄▃ beats 6/146easy · avg 83%▇▇ 4.0%at its level · noisy
EuroEval HellaSwag (English), test split, 5-shot generative MC, accuracy32.1% ✎▃·▁▁▃▁▂▅█▇▄▁ beats 14/100medium · avg 68%▇ 3.5%at its level · noisy
Chess Puzzles (Epoch)0.0% ✎█▁▁▁▁▂▁▄▂▂▁▁ beats 25/135very hard · avg 5.9%▇ 0.2%sharp · too hard for it · 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 Qwen 1.5B · needs ~1.1 GB · from huggingface.co/lmstudio-community
    chat in the terminal
    llama cli -hf lmstudio-community/DeepSeek-R1-Distill-Qwen-1.5B-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-Qwen-1.5B-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 Qwen 1.5B · needs ~1.1 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-Qwen-1.5B-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 Qwen 1.5B · needs ~1.57 GB · from ollama.com
    chat in the terminal
    ollama run deepseek-r1:1.5b

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 Qwen 1.5B · needs ~1 GB, Mac only · from huggingface.co/mlx-community
    chat in the terminal
    mlx_lm.chat --model mlx-community/DeepSeek-R1-Distill-Qwen-1.5B-4bit
  3. Or start an OpenAI-compatible server at http://localhost:8080/v1
    mlx_lm.server --model mlx-community/DeepSeek-R1-Distill-Qwen-1.5B-4bit

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

$ how the score was worked out · deepseek-ai/DeepSeek-R1 model… 49%, CodeReviewQA: The Code Review… 11%, BRIDGE Medical Leaderboard (H… 10% 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▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 49%mit
CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)4▇▇▇ 11%CC-BY-4.0
BRIDGE Medical Leaderboard (Harvard Medical School / Brigham and Women's YLab)1▇▇▇ 10%MIT
Epoch AI — internal evaluations3▇▇ 8%CC BY 4.0
THINK-Bench: Evaluating Thinking Efficiency and Chain-of-Thought Quality of Large Reasoning Models (arXiv:2505.22113v1)1▇▇ 8%CC-BY-4.0
EuroEval leaderboards (results collection), English generative leaderboard2▇▇ 7%MIT
Open Portuguese LLM Leaderboard (archived)1▇▇ 5%Apache-2.0
prior · pull toward the middle▇ 1%
weighted average= 37.2
$ quantization evidence · 14 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
FP8 W8A8 dynamic (FP8 weights symmetric per-channel, FP8 activations dynamic per-token; llm-compressor FP8_DYNAMIC)GPQA Diamond pass@10.354 → 0.359101.6%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)0.346 → 0.349100.8%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)0.5 → 0.502100.3%source ↗
FP8 W8A8 dynamic (FP8 weights symmetric per-channel, FP8 activations dynamic per-token; llm-compressor FP8_DYNAMIC)MATH-500 pass@10.847 → 0.847100.1%source ↗
FP8 W8A8 dynamic (FP8 weights symmetric per-channel, FP8 activations dynamic per-token; llm-compressor FP8_DYNAMIC)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)0.48 → 0.47899.6%source ↗
FP8 W8A8 dynamic (FP8 weights symmetric per-channel, FP8 activations dynamic per-token; llm-compressor FP8_DYNAMIC)MMLU 5-shot acc (lm-evaluation-harness, vLLM)0.374 → 0.37299.6%source ↗
FP8 W8A8 dynamic (FP8 weights symmetric per-channel, FP8 activations dynamic per-token; llm-compressor FP8_DYNAMIC)AIME 2024 pass@10.3 → 0.29899.3%source ↗
FP8 W8A8 dynamic (FP8 weights symmetric per-channel, FP8 activations dynamic per-token; llm-compressor FP8_DYNAMIC)GSM8K 5-shot strict-match (lm-evaluation-harness, vLLM)0.7 → 0.6998.6%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)0.48 → 0.47699.1%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)0.374 → 0.3798.9%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)0.7 → 0.6897.2%source ↗
INT4 W4A16 (weights only, symmetric per-group, group size 128; card states GPTQ via llm-compressor)GPQA Diamond pass@10.354 → 0.33293.8%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)0.5 → 0.46893.5%source ↗
INT4 W4A16 (weights only, symmetric per-group, group size 128; card states GPTQ via llm-compressor)AIME 2024 pass@10.3 → 0.24682.0%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