whatmodel.sh / DeepSeek-V2 Lite Chat (16B-A2.4B)

DeepSeek-V2 Lite Chat (16B-A2.4B)

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

DeepSeek-V2 Lite Chat (16B-A2.4B) is an open-weights language model from DeepSeek, released May 2024. Its published size is 15.71B parameters (2.4B active per token) and a 32k-token context window. Licence: deepseek.

Catalogue identity notes

Maker repo deepseek-ai/DeepSeek-V2-Lite-Chat @8586474: licence tag other (deepseek). 15,706,484,224 parameters from the Hugging Face safetensors metadata. Mixture-of-experts: pinned README: 'DeepSeek-V2-Lite comprises 15.7B total parameters, of which 2.4B are activated for each token'. Released 2024-05: month the maker's Hugging Face repo was created (createdAt 2024-05-15). Ollama tags deepseek-v2:16b-lite-chat-q4_K_M, deepseek-v2:16b-lite-chat-q8_0 are the library's instruct/chat builds of this checkpoint (tag names on the pinned tags page); download sizes are estimated from the parameter count.

38.0±1.2#565 of 719 LLMsfrom 15 benchmark results across 5 sources

$ how it compares

The LLMs ranked either side of DeepSeek-V2 Lite Chat (16B-A2.4B), out of 828 with a score, then more from DeepSeek.

#modelscorewhereRAMreleased
656LLaMA 65B Meta38±1comparison only2023-02
657Granite 3B Code Instruct (2K) IBM38±2local~3.7GB2024-04
658DeepSeek-V2 Lite Chat (16B-A2.4B) DeepSeek38±1local~11GB2024-05
659BioMistral-7B BioMistral38±3local~6.1GB2024-02
660Qwen3.5 0.8B Alibaba38±1local~2GB2026-03
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▁▁▁▃▆██▇▆▄▃▁ #382 of 458 · 29% of its score · 7 tests

Beats DeepSeek R1 Distill Qwen 7B, Qwen3.5 2B, DeepSeek R1 Distill Llama 8B here, though they rank higher overall. Trails Qwen2.5 1.5B Instruct, WizardLM 13B V1.2, Falcon3 3B Instruct here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
LiveCodeBench official (Jan 1–Nov 1, 2024)3.6% ✎▃▁·▁▂▄▆█▅▂·▂ beats 0/31hard · avg 26%▇▇▇▇ 11%sharp
HumanEval lab57.3% ✎▁▁▁▁▁▃▅▅█▃▂▁ beats 96/190easy · avg 84%▇▇▇ 8.6%sharp · at its level · small run
MBPP lab45.8% ✎▁▁▁▄▅█▄▅▃▁▂▁ beats 23/81easy · avg 71%▇▇ 4.3%at its level · small run
CodeReviewQA: solution identification (hard), invariant accuracy20.5% ✎▃··▁▂▂▃▂█▃▄▂ beats 15/47medium · avg 54%▇ 2.2%sharp · noisy
CodeReviewQA: automated code refinement, exact match rate15.0% ✎▁▁▁▁▁▁▃▂▃█▅▄ beats 10/47hard · avg 33%▇ 1.7%typical
CodeReviewQA: change localisation (hard), invariant accuracy6.9% ✎█▂▁▁▂▃▂▃▅▅▅▇ beats 16/47medium · avg 36%▇ 1.5%sharp · noisy
CodeReviewQA: change type recognition, invariant accuracy76.8% ✎▁▁▁······▁▃█ beats 38/47easy · avg 67%▇ 0.1%blunt · at its level · noisy
math█··········█ #1 of 2 * · 12% of its score · 2 tests
testresultvs the others in its rundifficultycountswhy
GSM8K lab72.0% ✎▂▂▂▅▅▆▅▇█▇▄▃ beats 73/146very easy · avg 89%▇▇▇ 7.2%sharp · at its level · small run
MATH lab27.9% ✎▃▅▂▆▅▃▅▃█▃▃▂ beats 16/29medium · avg 47%▇▇ 4.9%small run
other tests6 not in a task · 58% of its score
testresultvs the others in its rundifficultycountswhy
LiveBench 2024-07-2617.7% ✎▁▁▁▂▃▃▂█▆▃▅▃ beats 11/61medium · avg 41%▇▇▇▇▇▇▇▇▇▇▇▇▇ 34%typical
BIG-Bench Hard lab48.1% ✎▁▁▁▁▂▄▅▅█▃▄▁ beats 41/129easy · avg 75%▇▇▇ 8.3%sharp · at its level · small run
MMLU lab55.7% ✎▁▁▁▁▁▂▃▅█▃▆▃ beats 56/183easy · avg 77%▇▇ 5.9%at its level · small run
C-Eval lab60.1% ✎▁···▁▁▂▁▃▂▃█ beats 8/33easy · avg 74%▇▇ 4.3%at its level · small run
CMMLU lab62.5% ✎▅·▅····█▃▅█▅ beats 3/14easy · avg 76%▇▇ 3.8%at its level · small run
Open PT-LLM Leaderboard average (9 Portuguese tasks)63.3% ✎▃▁▁▁▁▁▁▁▃█▅▁ beats 102/276medium · avg 73%▇ 2.5%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-V2 Lite Chat (16B-A2.4B) · needs ~10.4 GB · from huggingface.co/mradermacher
    chat in the terminal
    llama cli -hf mradermacher/DeepSeek-V2-Lite-Chat-GGUF:Q4_K_M
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf mradermacher/DeepSeek-V2-Lite-Chat-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-V2 Lite Chat (16B-A2.4B) · needs ~10.4 GB · from huggingface.co/mradermacher

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/mradermacher/DeepSeek-V2-Lite-Chat-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-V2 Lite Chat (16B-A2.4B) · needs ~9.93 GB · from ollama.com
    chat in the terminal
    ollama run deepseek-v2:16b-lite-chat-q4_K_M

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-V2 Lite Chat (16B-A2.4B) · needs ~8.8 GB, Mac only · from huggingface.co/mlx-community
    chat in the terminal
    mlx_lm.chat --model mlx-community/DeepSeek-V2-Lite-Chat-4bit-mlx
  3. Or start an OpenAI-compatible server at http://localhost:8080/v1
    mlx_lm.server --model mlx-community/DeepSeek-V2-Lite-Chat-4bit-mlx

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

$ how the score was worked out · DeepSeek vendor-reported benc… 47%, LiveBench model judgments (HF… 34%, LiveCodeBench official websit… 11% and 2 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 vendor-reported benchmark scores for its own models (Hugging Face model cards) · lab-reported8▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 47%Factual vendor-reported score (not licensed content)
LiveBench model judgments (HF livebench/model_judgment, 2024-10-22 revision) with the LiveBench question datasets1▇▇▇▇▇▇▇▇▇▇ 34%Apache-2.0 (stated in the LiveBench datasheet)
LiveCodeBench official website results (2024-12-26)1▇▇▇ 11%CC-BY-SA-4.0
CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)4▇▇ 6%CC-BY-4.0
Open Portuguese LLM Leaderboard (archived)1▇ 2%Apache-2.0
prior · pull toward the middle▇ 1%
weighted average= 38.0

$ 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