whatmodel.sh / Gemma 2 2B IT

Gemma 2 2B IT

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

Gemma 2 2B IT is an open-weights language model from Google, released July 2024. Its published size is 2.61B parameters and a 8k-token context window. Licence: gemma.

Catalogue identity notes

Arena entry 'gemma-2-2b-it' (text). LMArena's leaderboard table (arena-leaderboard space @6f2eeeb) labels 'gemma-2-2b-it' as 'Gemma-2-2b-it' (Google, Gemma license), linking https://aistudio.google.com/app/prompts/new_chat?model=gemma-2-2b-it. Maker repo google/gemma-2-2b-it @299a856: licence tag gemma. 2,614,341,888 parameters from the Hugging Face safetensors metadata (dense, so active = total). Released 2024-07: Google Gemma releases page (ai.google.dev/gemma/docs/releases): 'July 31, 2024 Release of Gemma 2 in 2B size'.

37.2±0.6#575 of 719 LLMsfrom 29 benchmark results across 10 sources

$ how it compares

The LLMs ranked either side of Gemma 2 2B IT, out of 828 with a score, then more from Google.

#modelscorewhereRAMreleased
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
670Granite 20B Code Base (8K) IBM37±1local~15GB2024-04
more from Google
9Gemini 4 Argon (limited preview) Google82±1limited preview2026-09
23Gemini 3.8 Flash Google77±1api $1.50/M2026-09
32Gemini 3.7 Flash Google76±1api $1.50/M2026-08

see all LLMs · find the ones that fit your machine

$ what it's good at

coding▁▁▁▃▆██▇▆▄▃▁ #403 of 458 · 15% of its score · 3 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, Code Llama 7B Instruct here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
CRUXEval (ZeroEval, 800 problems)21.5% ✎▃▂▆▃▆█▆▃▂·▁▁ beats 0/47medium · avg 61%▇▇▇ 7.9%sharp
HumanEval40.6% ✎▄·▄█·▄·▄···█ beats 3/7easy · avg 81%▇▇ 4.8%sharp · at its level
MBPP34.0% ✎▄···▄·▄▄█··█ beats 2/7medium · avg 64%▇ 2.3%at its level
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #305 of 393 * · 5.1% of its score · 2 tests

Beats Gemma 4 E4B, OpenThinker3 7B, Gemma 4 E2B here, though they rank higher overall. Trails LFM2 350M, Gemma 3 1B, Flan-T5 XXL here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
LMArena (text)1200.0 ✎▁▁▁▂▃▃▃▆▆▇█▃ beats 69/379avg 1268▇ 3.7%blunt
IFEval (aggregation unspecified)53.8% ✎▄····█·▄··██ beats 7/7easy · avg 68%▇ 1.4%at its level
math▁▃▃▅▆▅▄█▄▁▁▁ #384 of 398 · 19% of its score · 4 tests

Beats Command R (03-2024), Qwen2-VL 2B Instruct here, though they rank higher overall. Trails Qwen2 0.5B Instruct, Granite 3.1 1B-A400M Instruct, Gemma 2B (base) here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
GSM8K (ZeroEval, zero-shot)51.6% ✎▁▄·▁▄█▅▅▅███ beats 1/43very easy · avg 89%▇▇▇▇ 10%sharp · at its level
MATH (subset unspecified)21.8% ✎▄···▄▄▄▄··█▄ beats 3/7medium · avg 53%▇ 3.6%typical
GSM8K30.9% ✎█▄····▄··█▄▄ beats 2/7easy · avg 81%▇ 3.6%sharp · at its level
MATH Level 5 (ZeroEval, 721 problems)4.3% ✎▃▁▁▆█▂▇▇▄··▂ beats 0/38hard · avg 34%▇ 1.6%sharp · too hard for it
other tests20 not in a task · 61% of its score
testresultvs the others in its rundifficultycountswhy
MMLU-Redux (ZeroEval, 2,778 questions)51.9% ✎▁·▃▁▄▄▄▅█▄▂▁ beats 1/53easy · avg 77%▇▇▇▇▇▇▇ 18%at its level
BIG-Bench Hard43.5% ✎▃▃···▃··█··▅ beats 3/7medium · avg 67%▇▇▇ 6.8%sharp
WinoGrande68.9% ✎█···▄▄▄▄···█ beats 3/7easy · avg 84%▇▇ 4.3%at its level
WildBench v2 WB-Score (task macro)27.8260 WB-Score ✎▁·▁▁·▃█▄▇▆▅▆ beats 9/52avg 45.199 WB-Score▇▇ 4.2%sharp
MMLU56.8% ✎▄····▄·▄▄█·█ beats 4/7easy · avg 77%▇ 3.7%at its level
GPQA (subset unspecified)29.4% ✎▄·▄··▄··▄█·█ beats 5/7hard · avg 39%▇ 3.5%sharp · too hard for it
MMLU-Pro27.2% ✎▄·····█··█▄█ beats 3/7medium · avg 47%▇ 2.8%typical
CPTUB average (SpeakLeash, 0-5 judge scale)53.0% ✎▁▂▂▂▃█▃▆▅▄▂▃ beats 19/95easy · avg 66%▇ 2.7%at its level
12 more tests
testresultvs the others in its rundifficultycountswhy
HellaSwag71.2% ✎▄·▄··█▄·▄·▄▄ beats 2/7easy · avg 87%▇ 2.5%at its level
BoolQ84.4% ✎▄▄····▄▄·█·█ beats 4/7easy · avg 91%▇ 2.4%at its level
ARC Challenge57.4% ✎▄····█··▄▄▄█ beats 5/7easy · avg 78%▇ 1.9%sharp · at its level · noisy
OpenBookQA44.2% ✎▄▄··▄··▄·█▄▄ beats 6/7medium · avg 60%▇ 1.8%typical
Open Persian LLM Leaderboard v2 average (21 tasks)44.4% ✎▂▄▂▃█▄▃▆▃▅▂▂ beats 18/81medium · avg 58%▇ 1.6%blunt · at its level
ZebraLogic (ZeroEval, 1,000 puzzles)4.2% ✎▁▃▃▆██▁▁·▁▁▁ beats 1/57hard · avg 18%▇ 1.2%sharp · too hard for it
Open PT-LLM Leaderboard average (9 Portuguese tasks)67.7% ✎▃▁▁▁▁▁▁▁▂█▅▁ beats 136/276medium · avg 73%▇ 0.7%at its level · noisy
SurgiQ surgical MCQ (zero-shot, log-probability scoring)35.0% ✎▂···▂▃·▃▂▃██ beats 2/17medium · avg 53%▇ 0.6%typical
EuroEval Life in the UK (English knowledge MC), test split, 5-shot, accuracy70.0% ✎▁▁▁▁▁▁▁▂▄█▄▃ beats 38/146easy · avg 83%▇ 0.6%at its level · noisy
EuroEval HellaSwag (English), test split, 5-shot generative MC, accuracy53.9% ✎▃·▁▁▂▁▂▅█▇▄▁ beats 29/100medium · avg 68%▇ 0.5%at its level · noisy
Social IQA60.8% ✎▃·····▃··█▃▅ beats 4/7medium · avg 68%▇ 0.5%blunt · at its level
Vectara hallucination leaderboard, 2025 dataset (HHEM-2.3, factual consistency rate)93.0% ✎▁▁▂▃▄▄█▆▃▂▂▁ beats 43/140very easy · avg 94%▇ 0.4%blunt · 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 Gemma 2 2B IT · needs ~1.7 GB · from huggingface.co/lmstudio-community
    chat in the terminal
    llama cli -hf lmstudio-community/gemma-2-2b-it-GGUF:Q4_K_M
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf lmstudio-community/gemma-2-2b-it-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 Gemma 2 2B IT · needs ~1.7 GB · from huggingface.co/lmstudio-community

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/lmstudio-community/gemma-2-2b-it-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 Gemma 2 2B IT · needs ~2.07 GB · from ollama.com
    chat in the terminal
    ollama run gemma2:2b-instruct-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 Gemma 2 2B IT · needs ~1.5 GB, Mac only · from huggingface.co/mlx-community
    chat in the terminal
    mlx_lm.chat --model mlx-community/gemma-2-2b-it-4bit
  3. Or start an OpenAI-compatible server at http://localhost:8080/v1
    mlx_lm.server --model mlx-community/gemma-2-2b-it-4bit

Official weights and documentation: huggingface.co/google ↗

$ how the score was worked out · IBM Granite 3.0 Language Mode… 46%, WildEval ZeroEval published r… 39%, WildBench v2 leaderboard data… 4% and 7 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
IBM Granite 3.0 Language Models technical report (GitHub)15▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 46%Apache-2.0
WildEval ZeroEval published results5▇▇▇▇▇▇▇▇▇▇▇▇ 39%Apache-2.0
WildBench v2 leaderboard data (Ai2)1▇ 4%Apache-2.0
LMArena leaderboard dataset1▇ 4%CC BY 4.0
SpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)1▇ 3%MIT
Open Persian LLM Leaderboard v2 (Part DP AI and AUT NLP Lab)1▇ 2%Apache-2.0
EuroEval leaderboards (results collection), English generative leaderboard2▇ 1%MIT
Open Portuguese LLM Leaderboard (archived)1▇ 1%Apache-2.0
SurgiQ: A Large-Scale Multi-Domain Benchmark for Evaluating Surgical Understanding in Large Language Models (arXiv:2606.08071v1)1▇ 1%CC BY 4.0
Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)1▇ 0%Apache-2.0
prior · pull toward the middle▇ 0%
weighted average= 37.2

$ 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