whatmodel.sh / DeepSeek V3 0324

DeepSeek V3 0324

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

DeepSeek V3 0324 is an open-weights language model from DeepSeek, released March 2025. Its published size is 671B parameters (37B active per token) and a 164k-token context window. Licence: mit.

60.3±0.7#192 of 719 LLMsfrom 30 benchmark results across 17 sources

$ how it compares

The LLMs ranked either side of DeepSeek V3 0324, out of 828 with a score, then more from DeepSeek.

#modelscorewhereRAMreleased
228Mellum2.1 12B-A2.5B Thinking JetBrains60±2local~8.3GB2026-09
229GPT-4.1 mini OpenAI60±1api $0.70/M2025-04
230DeepSeek V3 0324 DeepSeek60±1local~108GB2025-03
231Qwen3 30B-A3B Instruct 2507 Alibaba60±1api $0.15/M +local~20GB2025-07
232Qwen3-VL 32B Instruct Alibaba60±1api $0.28/M +local~22GB2025-10
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

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #173 of 300 · 22% of its score · 7 tests

Beats o1 (2024-12-17), o3-mini, DeepSeek R1 0528 here, though they rank higher overall. Trails SmolLM3 3B, Jamba 1.5 Mini, Jamba 1.5 Large here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
τ²-bench Telecom +132.5% ✎██··█·██··██ beats 2/6very hard · avg 10%▇▇▇▇▇▇ 15%sharp · at its level · several runs
SWE-bench Verified lab45.4% ✎▁▁▁▁▁▁▂▄█▆▁▁ beats 33/131hard · avg 22%▇ 2.7%sharp · at its level · small run
BFCL V4 (2025-08-26 archive, models absent from later archives)45.2% ✎▄▄····█▄···▄ beats 3/5hard · avg 27%▇ 1.6%at its level
SWE-bench Multilingual lab29.3% ✎▁··▁▃▂▅▇█▃▃▁ beats 4/60hard · avg 19%▇ 1.5%at its level · small run
BFCL v364.7% ✎▁▁▁▁▁▃▅█▅▄▁▁ beats 96/114medium · avg 52%▇ 0.9%blunt · at its level
Galileo Agent Leaderboard v1 (tool selection quality)90.5% ✎▃▃▃▃▆▅▆▅█▂▅▂ beats 24/31easy · avg 79%▇ 0.6%blunt
Terminal-Bench (pre-2.0) with Terminus 1 lab13.3% ✎█····█·····█ beats 1/2very hard · avg 7.1%▇ 0.4%noisy · small run
coding▁▁▁▃▆██▇▆▄▃▁ #144 of 458 · 32% of its score · 6 tests

Beats Gemini 2.5 Pro, Gemini 3.1 Flash-Lite, Qwen3-Coder-Next (80B-A3B) here, though they rank higher overall. Trails Phi-4-reasoning 14B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Aider polyglot (225 exercises, two attempts) +155.1% ✎▂·▂▃▅▂▃█▆▄▂▂ beats 31/49very hard · avg 8.5%▇▇▇▇▇▇▇▇▇ 23%sharp · at its level · several runs
SWE-bench Verified lab45.4% ✎▁▁▁▁▁▁▂▄█▆▁▁ beats 33/131hard · avg 22%▇ 2.7%sharp · at its level · small run
WebApp1K-Duo-React pass@1 (onekq-ai)61.6% ✎▁··▁·▁▁▁▆█▇█ beats 17/45hard · avg 32%▇ 2.3%at its level
LiveCodeBench (version unspecified) lab49.2% ✎▁···▁▁▃▅▅█▄▅ beats 17/37hard · avg 30%▇ 1.6%at its level · small run
SWE-bench Multilingual lab29.3% ✎▁··▁▃▂▅▇█▃▃▁ beats 4/60hard · avg 19%▇ 1.5%at its level · small run
LiveCodeBench 2024-08–2025-05 (version unstated) lab43.0% ✎▃···█··▅··▃▃ beats 0/7medium · avg 40%▇ 1.1%at its level · small run
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #116 of 393 * · 6.3% of its score · 2 tests

Beats LongCat 2.0 (1.6T-A48B), Step 3.5 Flash (196B-A11B), GPT-5 mini here, though they rank higher overall. Trails LFM2.5 2.6B, Nemotron-Cascade-8B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
LMArena (text)1395.6 ✎▁▁▁▂▃▃▃▆▆▇█▃ beats 240/379avg 1268▇▇ 4.5%blunt
IFEval prompt-level strict81.9% ✎▄▄··▄·▄··█·▄ beats 2/6easy · avg 74%▇ 1.8%typical
math▁▂▃▅▆▅▄█▄▁▁▁ #190 of 398 · 18% of its score · 6 tests

Beats MiniMax-M2.1 here, though they rank higher overall. Trails Phi-4-mini-reasoning 3.8B, DeepSeek R1 Distill Qwen 7B, OpenThinker3 7B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
MATH level 575.5% ✎▂▂▅▆▄█▆▃▅▂▃▄ beats 64/92hard · avg 32%▇▇▇ 7.6%sharp
OTIS mock AIME 2024–202537.8% ✎▃▃▄▃▃▃▇▇█▃▂▅ beats 70/193very hard · avg 8.5%▇▇ 4.5%sharp · at its level · noisy
AIME 2025 lab51.3% ✎▁·▁▁▁▂▅▆█▇▅▁ beats 38/133medium · avg 36%▇ 1.9%sharp · at its level · noisy · small run
AIME 2024 lab59.4% ✎▁▁▁▂▁▃▁▅███▃ beats 28/85medium · avg 47%▇ 1.8%sharp · noisy · small run
HMMT 2025 (edition unspecified) lab29.2% ✎▃█▇▅▇▅▁·▁··▁ beats 0/32hard · avg 24%▇ 1.7%sharp · at its level · noisy · small run
MATH-500 lab94.0% ✎▁▃▂▃▅▇▅▇█▅▂▁ beats 48/84very easy · avg 89%—not counted yet: no other model in the same run to compare it with
other tests9 not in a task · 25% of its score
testresultvs the others in its rundifficultycountswhy
GPQA Diamond +167.6% ✎▁▁▁▁▂▄▆▄▅▇█▅ beats 102/208hard · avg 40%▇▇▇▇▇ 12%sharp · at its level · several runs
EQ-Bench 3 (emotional intelligence, judge Elo)1172.2 ✎▁·▂▁▄▅▅▃█▄▁▁ beats 27/62avg 1044▇ 2.7%typical
MMLU-Redux lab90.5% ✎▁▁▃▁▂▃▄▅▇▅█▄ beats 38/63easy · avg 80%▇ 2.4%small run
CPTUB average (SpeakLeash, 0-5 judge scale)80.6% ✎▁▂▂▂▃█▃▆▅▄▂▃ beats 84/95easy · avg 66%▇ 2.2%typical
MMLU-Pro lab81.2% ✎▁▁·▁▁▁▃▂▄▅█▁ beats 133/188medium · avg 60%▇ 1.9%small run
Open Persian LLM Leaderboard v2 average (21 tasks)72.4% ✎▂▄▂▃█▄▃▆▃▅▃▂ beats 76/81medium · avg 58%▇ 1.7%blunt · at its level
Polish EQ-Bench v2 (SpeakLeash)73.5% ✎▁··▁▁▂▂▆▅▆█▆ beats 64/70medium · avg 64%▇ 1.6%typical
Open PT-LLM Leaderboard average (9 Portuguese tasks)81.5% ✎▃▁▁▁▁▁▁▁▂█▅▁ beats 265/276medium · avg 73%▇ 0.7%noisy
1 more test
testresultvs the others in its rundifficultycountswhy
Vectara hallucination leaderboard, 2025 dataset (HHEM-2.3, factual consistency rate)92.0% ✎▁▁▂▃▄▄█▆▃▂▂▁ beats 36/140very easy · avg 94%▇ 0.2%blunt · maxed out · 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 V3 0324 · needs ~403.1 GB · from huggingface.co/unsloth
    chat in the terminal
    llama cli -hf unsloth/DeepSeek-V3-0324-GGUF:UD-Q4_K_XL
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf unsloth/DeepSeek-V3-0324-GGUF:UD-Q4_K_XL

LM Studio

version

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

    download it (macOS / Windows / Linux) ↗

  2. Run DeepSeek V3 0324 · needs ~403.1 GB · from huggingface.co/unsloth

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/unsloth/DeepSeek-V3-0324-GGUF@UD-Q4_K_XL -y
  4. Then run lms ls to see its name, and lms chat followed by that name

MLX (Mac)

  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 V3 0324 · needs ~461.5 GB, Mac only · from huggingface.co/mlx-community
    chat in the terminal
    mlx_lm.chat --model mlx-community/DeepSeek-V3-0324-5bit
  3. Or start an OpenAI-compatible server at http://localhost:8080/v1
    mlx_lm.server --model mlx-community/DeepSeek-V3-0324-5bit

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

$ how the score was worked out · deepseek-ai/DeepSeek-V3.1 mod… 30%, Epoch AI — internal evaluatio… 20%, Aider polyglot benchmark runs 13% and 14 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-V3.1 model card · lab-reported11▇▇▇▇▇▇▇▇▇ 30%MIT
Epoch AI — internal evaluations3▇▇▇▇▇▇ 20%CC BY 4.0
Aider polyglot benchmark runs1▇▇▇▇ 13%Apache-2.0
MUA-RL: Multi-turn User-interacting Agent Reinforcement Learning for agentic tool use (arXiv:2508.18669v1)1▇▇▇ 11%CC-BY-4.0
LMArena leaderboard dataset1▇ 4%CC BY 4.0
Moonshot AI Kimi-K2-Instruct model card (instruction-model evaluation table)1▇ 4%Modified MIT License (Kimi K2)
EQ-Bench 3 canonical leaderboard results1▇ 3%MIT
Berkeley Function Calling Leaderboard result archive2▇ 3%Apache-2.0
WebApp1K models leaderboard (onekq-ai)1▇ 2%MIT
SpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)1▇ 2%MIT
OpenCompass CompassAcademic leaderboard (full version) Space: committed predictions file1▇ 2%Apache-2.0
Open Persian LLM Leaderboard v2 (Part DP AI and AUT NLP Lab)1▇ 2%Apache-2.0
SpeakLeash Polish EQ-Bench leaderboard1▇ 2%Apache-2.0
deepseek-ai/DeepSeek-V3-0324 model card · lab-reported1▇ 2%MIT
Open Portuguese LLM Leaderboard (archived)1▇ 1%Apache-2.0
Galileo Agent Leaderboard (v1 and v2 results)1▇ 1%Apache-2.0
Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)1▇ 0%Apache-2.0
prior · pull toward the middle▇ 0%
weighted average= 60.3

not counted (1)

deepseek-ai/DeepSeek-V3-0324 model cardMATH-500 [correct]94.0%fewer than three models have this test yet, or this source ran it for this model alone

$ 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