whatmodel.sh / Gemma 2B (base)

Gemma 2B (base)

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

Gemma 2B (base) is an open-weights language model from Google, released February 2024. Its published size is 2.51B parameters and a 8k-token context window. Licence: gemma.

Catalogue identity notes

Maker repo google/gemma-2b @9cf48e5: licence tag gemma. 2,506,172,416 parameters from the Hugging Face safetensors metadata (dense, so active = total). Released 2024-02: month the maker's Hugging Face repo was created (createdAt 2024-02-08). Pretrained Gemma 1.0 2B; the catalogue's gemma-2b is the instruction-tuned Gemma 2B IT. Ollama tags gemma:2b-text-q4_K_M, gemma:2b-text-q8_0 (sizes as listed on the tags page, observed 2026-09-29).

28.9±0.6#648 of 719 LLMsfrom 37 benchmark results across 9 sources

$ how it compares

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

#modelscorewhereRAMreleased
745OLMo 7B Instruct Ai229±2local~16GB2024-02
746StableLM-3B-4E1T Stability AI29±1local~3.3GB2023-09
747Gemma 2B (base) Google29±1local~3.2GB2024-02
748OpenLLaMA 7B v2 OpenLM Research29±1local~9.1GB2023-07
749Gemma 1.1 2B IT Google29±1local~3.2GB2024-04
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▁▁▁▃▆██▇▆▄▃▁ #438 of 458 · 66% of its score · 7 tests

Beats Qwen3.5 2B, Molmo 7B-D (0924), DeepSeek R1 Distill Qwen 1.5B here, though they rank higher overall. Trails Falcon3 1B Instruct, Granite 3.0 1B-A400M Instruct here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
HumanEval +325.0% ✎▁·▁▁▃▄▆▅█▃▁▁ beats 9/99easy · avg 79%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 38%sharp · several runs
HumanEval+20.7% ✎▁·▁▁▁▃▄▄▄█▃▁ beats 8/99easy · avg 74%▇▇▇▇▇▇ 14%sharp
HumanEvalSynthesize (HumanEvalPack, 6-language average)17.2% ✎▁·▃▁▂▆▅█▄▄▁▂ beats 1/35medium · avg 64%▇▇ 4.3%typical
MBPP+34.1% ✎▁···▁▃▂▅▇█▆▂ beats 2/75easy · avg 68%▇ 3.7%at its level
MBPP lab +129.2% ✎▁▁▃▄▅█▄▅▃▁▂▁ beats 5/81easy · avg 74%▇ 3.0%at its level · several runs
HumanEvalExplain (HumanEvalPack, 6-language average)10.9% ✎▄▇█▃▄▇▄▁▇▃·▁ beats 1/35medium · avg 49%▇ 2.0%typical
HumanEvalFix (HumanEvalPack, 6-language average)1.9% ✎▃·▁▁▄██▄█▁▅▅ beats 0/35medium · avg 56%▇ 1.4%sharp
math▁▂▃▅▆▅▄█▄▁▁▁ #373 of 398 * · 3.2% of its score · 2 tests

Beats Command R (03-2024), Molmo 7B-D (0924), Mistral 7B Instruct v0.3 here, though they rank higher overall.

testresultvs the others in its rundifficultycountswhy
GSM8K lab17.7% ✎▂▂▃▅▅▆▅▇█▇▄▃ beats 10/146easy · avg 79%▇ 2.0%sharp · at its level · small run
MATH lab11.8% ✎▃▅▂▆▅▃▅▃█▃▃▂ beats 9/29medium · avg 50%▇ 1.2%small run
other tests22 not in a task · 30% of its score
testresultvs the others in its rundifficultycountswhy
EvoEval (overall)12.8% ✎▃▂▃▄▄█▁▄▅▃▁▃ beats 0/42medium · avg 55%▇▇ 5.3%typical
Natural Questions (closed book) +114.8% ✎▁·▁·▁▂▁▂·▅▂█ beats 13/28medium · avg 58%▇▇ 4.8%several runs
MedQA (USMLE, 4 options)29.2% ✎▁▁··▄▃▁··▄▃█ beats 5/17easy · avg 76%▇ 2.2%sharp
ARC Challenge lab42.1% ✎▆▁▁▂▂▅▇█▄▃▃▂ beats 43/97easy · avg 83%▇ 2.1%sharp · small run
BIG-Bench Hard lab35.2% ✎▁▁▁▁▂▄▅▅█▃▄▁ beats 15/129easy · avg 82%▇ 1.9%sharp · small run
TriviaQA +147.8% ✎▁··▁█▃▃▁▄▅▄█ beats 19/28very easy · avg 85%▇ 1.8%at its level · small run · several runs
MedMCQA30.2% ✎▂▂··▄▄▂▂·▄██ beats 2/17medium · avg 68%▇ 1.4%typical
WinoGrande lab65.4% ✎▁···▂▁▁▅█▇▃▁ beats 43/87easy · avg 87%▇ 1.3%at its level · small run
14 more tests
testresultvs the others in its rundifficultycountswhy
OpenBookQA lab47.8% ✎█▁▁▁▄▃▃▇▂▁▄▂ beats 42/65easy · avg 74%▇ 1.3%small run
BoolQ lab69.4% ✎▁▄▅▆▂▆▄▄█▃▄▄ beats 20/58easy · avg 92%▇ 1.3%at its level · small run
MMLU lab42.3% ✎▁▁▁▁▁▂▃▅█▃▆▃ beats 21/183easy · avg 81%▇ 1.2%at its level · small run
HellaSwag lab71.4% ✎▁▁▁▁▂▂▃▇█▆▅▃ beats 43/85very easy · avg 90%▇ 1.1%at its level · small run
AGIEval English (Gemma report) lab24.2% ✎▄··········█ beats 0/2easy · avg 65%▇ 0.9%at its level · small run
CommonsenseQA lab65.3% ✎▄·▄▄·▄····▄█ beats 4/6easy · avg 84%▇ 0.8%at its level · small run
ARC Easy lab73.2% ✎▁▂▃▂▃▅▇█▇▆▄▆ beats 50/65very easy · avg 90%▇ 0.7%at its level · small run
Social IQA lab49.7% ✎▂·▂██▅▅▁▂▃▆▁ beats 31/51medium · avg 65%▇ 0.5%blunt · small run
Open PT-LLM Leaderboard average (9 Portuguese tasks)44.1% ✎▃▁▁▁▁▁▃▁▂█▅▁ beats 54/276medium · avg 73%▇ 0.5%noisy
PIQA lab77.3% ✎▁▁▁▃▃▄▇█▄▃▁▁ beats 50/83easy · avg 87%▇ 0.4%blunt · at its level · small run
EuroEval HellaSwag (English), test split, 5-shot generative MC, accuracy26.9% ✎▃·▃▁▂▁▂▅█▇▄▁ beats 9/100medium · avg 68%▇ 0.3%noisy
TruthfulQA MC233.1% ✎▅▃█▆▁▂▅···▁▁ beats 0/31medium · avg 64%▇ 0.3%blunt · at its level
PubMedQA66.4% ✎▂·▂▆·▄▃▄▄▂█▂ beats 7/17medium · avg 77%▇ 0.2%blunt · at its level
HaluEval QA hallucination detection (accuracy)48.6% ✎█▁▁·▁▁·▁▁▁▁▁ beats 7/27medium · avg 70%▇ 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 Gemma 2B (base) · needs ~1.6 GB · from huggingface.co/MaziyarPanahi
    chat in the terminal
    llama cli -hf MaziyarPanahi/gemma-2b-GGUF:Q4_K_M
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf MaziyarPanahi/gemma-2b-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 2B (base) · needs ~1.6 GB · from huggingface.co/MaziyarPanahi

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/MaziyarPanahi/gemma-2b-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 2B (base) · needs ~1.8 GB · from ollama.com
    chat in the terminal
    ollama run gemma:2b-text-q4_K_M

Official weights and documentation: huggingface.co/google ↗

$ how the score was worked out · EvalPlus / EvalPerf published… 39%, Gemma report — original evalu… 22%, Top Leaderboard Ranking = Top… 16% and 6 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
EvalPlus / EvalPerf published evaluation runs3▇▇▇▇▇▇▇▇▇▇▇▇ 39%Apache-2.0
Gemma report — original evaluations · lab-reported18▇▇▇▇▇▇▇ 22%CC BY 4.0
Top Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runs2▇▇▇▇▇ 16%CC BY 4.0
Granite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runs3▇▇ 8%CC BY 4.0
CodeGemma: Open Code Models Based on Gemma — authors' own runs · lab-reported2▇▇ 5%CC BY 4.0
Hallucinations Leaderboard results (HF dataset hallucinations-leaderboard/results)4▇ 5%Apache-2.0
Open Medical-LLM Leaderboard results (HF dataset openlifescienceai/results)3▇ 4%Apache-2.0
Open Portuguese LLM Leaderboard (archived)1▇ 0%Apache-2.0
EuroEval leaderboards (results collection), English generative leaderboard1▇ 0%MIT
prior · pull toward the middle▇ 0%
weighted average= 28.9

$ 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