Gemma 4 E2B

+ contribute

Gemma 4 E2B is an open-weights language model from Google, released April 2026. Its published size is 5.1B parameters (2.3B active per token) and a 131k-token context window. Licence: apache-2.0.

46.7±0.6#404 of 719 LLMsfrom 78 benchmark results across 12 sources

$ how it compares

The LLMs ranked either side of Gemma 4 E2B, out of 828 with a score, then more from Google.

#	model	score	where	RAM	released
479	InternVL2-26B Shanghai AI Laboratory (OpenGVLab)	47±2	local	~55GB	2024-07
480	speechless-coder-ds-6.7b uukuguy	47±1	local	~5.8GB	2023-12
481	Gemma 4 E2B Google	47±1	api $0.050/M +local	~5.1GB	2026-04
482	Magicoder-S-CL-7B UIUC (ise-uiuc)	47±1	local	~5.8GB	2023-12
483	Aya Expanse 8B Cohere	47±2	local	~6.2GB	2024-10
	more from Google
9	Gemini 4 Argon (limited preview) Google	82±1	limited preview		2026-09
23	Gemini 3.8 Flash Google	77±1	api $1.50/M		2026-09
32	Gemini 3.7 Flash Google	76±1	api $1.50/M		2026-08

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #273 of 300 · 15% of its score · 15 tests

Beats Magistral Small 1.0 (2506), Bielik 11B v2.3 Instruct, Mixtral 8x22B Instruct v0.1 here, though they rank higher overall. Trails Granite 4.0 H 350M, Qwen3 0.6B, Qwen2.5 1.5B Instruct here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
BFCL V4 (December 2025 archive) +2	36.6% ✎	▄·█·▄▄▄▄·▄·▄ beats 0/8	medium · avg 43%	▇ 2.8%	at its level · several runs
Tau2 – airline lab	31.0% ✎	▃·▃··▃·····█ beats 0/5	medium · avg 42%	▇ 2.0%	at its level
τ²-bench aggregate lab	24.5% ✎	▄·▄··▄····█▄ beats 1/5	hard · avg 32%	▇ 1.4%	at its level
τ²-bench Telecom lab +1	19.7% ✎	▄····█··▄·▄▄ beats 2/5	very hard · avg 14%	▇ 1.3%	sharp · noisy · several runs
MCP-Atlas (public set)	12.6% ✎	▃·▁▂▁·▁▃█▆▆▂ beats 0/37	hard · avg 21%	▇ 1.2%	typical
ContractBench success rate (33 observation-contract tool tasks)	7.1% ✎	▆··▆▄▂▄▂▆▆█▄ beats 2/23	very hard · avg 14%	▇ 1.1%	sharp
τ³-bench Banking +1	3.9% ✎	▂▂·▅█▂▂▃▂▅▂▃ beats 3/20	very hard · avg 4.0%	▇ 1.0%	blunt · too hard for it · several runs
Terminal-Bench Hard lab	3.0% ✎	██··█·█·█··█ beats 0/5	very hard · avg 4.7%	▇ 0.8%	sharp · too hard for it

7 more tests

test	result	vs the others in its run	difficulty	counts	why
BrowseComp-ZH	4.7% ✎	▄▄·▄▄▄·█···█ beats 1/8	very hard · avg 13%	▇ 0.7%	typical
Tau2 – retail lab	34.6% ✎	▄····▄··▄·▄█ beats 1/5	medium · avg 40%	▇ 0.7%	sharp · at its level · noisy
SWE-bench Verified	2.0% ✎	▄··▄▄█··▄·▄█ beats 1/8	very hard · avg 12%	▇ 0.6%	sharp · noisy
τ³-bench	8.8% ✎	▃·▃·█·▃··▃·▃ beats 2/7	very hard · avg 14%	▇ 0.4%	blunt
SWE-bench Pro (split unspecified)	0.0% ✎	█▂···▂···▄·▂ beats 2/8	very hard · avg 4.3%	▇ 0.4%	sharp · too hard for it · noisy
BrowseComp	3.7% ✎	▅···█·▃···▃▃ beats 1/7	very hard · avg 10%	▇ 0.3%	noisy
T1-Bench multi-domain conversational agent pass^3	28.4% ✎	▂▂··▂·▂·▃▂█▄ beats 4/11	hard · avg 25%	▇ 0.0%	blunt · noisy

coding▁▁▁▃▆██▇▆▄▃▁ #349 of 458 · 10% of its score · 8 tests

Beats LFM2.5 2.6B, LFM2.5 8B-A1B, Qwen3-VL 2B Thinking here, though they rank higher overall. Trails Phi-3-mini-128k-instruct (Apr 2024), Code Tulu 2 13B, DeepSeek LLM 67B Chat here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LiveCodeBench v6 (window unspecified) +1	42.9% ✎	▃····█·▃▅▃·▃ beats 3/8	medium · avg 39%	▇ 3.7%	at its level · several runs
LiveCodeBench Pro 2025Q2 (Easy)	27.1% ✎	▄···█▄··▄█▄▄ beats 1/8	hard · avg 33%	▇ 1.7%	at its level
LiveCodeBench v6 (Feb–May 2025) lab	44.0% ✎	▄··▄·▄···▄·█ beats 1/5	medium · avg 43%	▇ 1.1%	at its level
SciCode lab	21.0% ✎	█▄·······▄▄▄ beats 1/5	hard · avg 21%	▇ 1.0%	typical
SWE-bench Verified	2.0% ✎	▄··▄▄█··▄·▄█ beats 1/8	very hard · avg 12%	▇ 0.6%	sharp · noisy
SciCode (level unspecified)	20.5% ✎	█·····██████ beats 2/6	hard · avg 17%	▇ 0.6%	typical
LiveCodeBench Pro 2025Q2 (Medium)	0.0% ✎	█···▂··▄▂··▂ beats 1/8	very hard · avg 2.0%	▇ 0.6%	sharp · too hard for it · noisy
SWE-bench Pro (split unspecified)	0.0% ✎	█▂···▂···▄·▂ beats 2/8	very hard · avg 4.3%	▇ 0.4%	sharp · too hard for it · noisy

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #367 of 393 · 1.8% of its score · 2 tests

Trails Gemma 3 270M IT, SmolLM2 360M Instruct, Granite 4.0 350M here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
IFEval (aggregation unspecified) +2	31.4% ✎	▄▄····▄··███ beats 0/8	easy · avg 79%	▇ 0.9%	several runs
IFEval lab	94.6% ✎	▄··▄··█···▄▄ beats 1/5	very easy · avg 94%	▇ 0.9%	maxed out

math▁▂▃▅▆▅▄█▄▁▁▁ #277 of 398 · 25% of its score · 6 tests

Beats Molmo 2 8B, DeepSeek-Coder-V2 Instruct (0724), Llama 3.1 Nemotron 70B Instruct here, though they rank higher overall. Trails DeepSeek R1 Distill Qwen 1.5B, Granite 8B Code Base (4K), Mixtral 8x7B v0.1 (base) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
HMMT February 2026 +1	17.8% ✎	▅··▃▃·▃··▃·█ beats 0/8	hard · avg 32%	▇▇▇ 7.9%	sharp · several runs
AIME 2025 +1	31.7% ✎	█▄▄▄▄····█·▄ beats 1/8	medium · avg 42%	▇▇ 4.9%	sharp · at its level · several runs
AIME 2026 +1	39.8% ✎	▄·▄█·▄▄···█▄ beats 1/8	medium · avg 48%	▇▇ 4.8%	sharp · at its level · several runs
MATH-500	85.4% ✎	██▄▄▄··▄···▄ beats 0/8	very easy · avg 89%	▇ 2.9%	typical
IMO-AnswerBench	22.6% ✎	▄▄···▄·▄··█▄ beats 0/6	hard · avg 32%	▇ 2.3%	at its level
MATH-Vision lab	52.4% ✎	██·█·····███ beats 1/5	medium · avg 53%	▇ 1.8%	at its level

vision & documents▁▁▃▂▁▃▄▂▆▇█▄ #132 of 147 · 12% of its score · 13 tests

Trails Qwen3-VL 2B Instruct, Ministral 3 3B, Qwen2-VL 7B Instruct here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
InfographicVQA lab	63.9% ✎	▄█·······█·▄ beats 0/5	easy · avg 71%	▇ 2.3%	at its level
MMMU Pro lab	44.2% ✎	██·█····█·██ beats 0/5	medium · avg 49%	▇ 1.9%	at its level
MATH-Vision lab	52.4% ✎	██·█·····███ beats 1/5	medium · avg 53%	▇ 1.8%	at its level
CharXiv descriptive questions	75.1% ✎	█·▄·······▄█ beats 3/5	easy · avg 78%	▇ 1.7%	at its level
MMStar	52.9% ✎	▄▄▄█·······▄ beats 3/5	medium · avg 62%	▇ 1.3%	at its level
OmniDocBench 1.5 ↓ lab	29.0% ✎	█·█·····████ beats 1/5	avg 31%	▇ 0.7%	blunt
MMBench-EN dev v1.1	69.3% ✎	█·····▃··▃·▃ beats 2/5	easy · avg 79%	▇ 0.7%	blunt
HallusionBench	59.8% ✎	▄▄·▄·····█·▄ beats 1/5	medium · avg 65%	▇ 0.6%	blunt · at its level

5 more tests

test	result	vs the others in its run	difficulty	counts	why
RealWorldQA	50.7% ✎	▄···█··▄▄··▄ beats 0/5	easy · avg 69%	▇ 0.3%	blunt · at its level
CountBench	53.4% ✎	▄··▄··█··▄·▄ beats 0/5	very easy · avg 86%	▇ 0.3%	noisy
OCRBench	71.9% ✎	█··███·█···█ beats 1/5	easy · avg 81%	▇ 0.2%	blunt
AI2D test	71.2% ✎	█···▄▄···▄·▄ beats 0/5	easy · avg 82%	▇ 0.1%	blunt · at its level
ChartQA	42.2% ✎	▄······▄··██ beats 0/5	easy · avg 80%	▇ 0.1%	blunt · noisy

web search█·█▄██·▄▄·▄▄ #12 of 13 * · 1.0% of its score · 2 tests

Beats Nemotron 3 Nano 4B here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
BrowseComp-ZH	4.7% ✎	▄▄·▄▄▄·█···█ beats 1/8	very hard · avg 13%	▇ 0.7%	typical
BrowseComp	3.7% ✎	▅···█·▃···▃▃ beats 1/7	very hard · avg 10%	▇ 0.3%	noisy

other tests21 not in a task · 39% of its score

test	result	vs the others in its run	difficulty	counts	why
MMLU-Pro +2	56.0% ✎	▃··▂█·▂▂···▂ beats 0/8	medium · avg 61%	▇▇▇ 6.6%	at its level · several runs
EuroEval Life in the UK (English knowledge MC), validation split, 5-shot, accuracy	69.7% ✎	▂·▃▂·▂·▃▃·█▃ beats 1/14	easy · avg 80%	▇ 3.2%	at its level
GPQA Diamond lab	43.4% ✎	█···▄·····▄█ beats 1/5	hard · avg 47%	▇ 3.2%	sharp
MMMLU lab	67.4% ✎	██··█···█·██ beats 0/5	medium · avg 69%	▇ 2.6%	at its level
IFBench (aggregation unspecified) +2	25.7% ✎	▅····▃▃·█·▃▃ beats 0/8	medium · avg 39%	▇ 2.4%	at its level · several runs
MMLU-Redux	71.8% ✎	▃···█··▂▂··▂ beats 0/8	easy · avg 77%	▇ 2.1%	at its level
GPQA (subset unspecified)	43.8% ✎	█···▃···▅·▃▃ beats 0/7	hard · avg 45%	▇ 1.9%	sharp
MMLU	69.2% ✎	▄·····▄▄··▄█ beats 5/5	medium · avg 70%	▇ 1.6%	at its level

13 more tests

test	result	vs the others in its run	difficulty	counts	why
IFBench lab	38.0% ✎	▄▄·▄······▄█ beats 1/5	medium · avg 40%	▇ 1.6%	at its level
Multi-IF (aggregation unspecified) +2	40.3% ✎	▄▄···▄·▄·▄██ beats 0/8	medium · avg 58%	▇ 1.6%	blunt · at its level · several runs
BRIDGE clinical text, zero-shot average (87 tasks)	32.7% ✎	▁▁▁▁▁▂▂▃▄▆█▃ beats 37/97	hard · avg 32%	▇ 1.5%	blunt · at its level
EuroEval HellaSwag (English), validation split, 5-shot generative MC, accuracy	53.6% ✎	▂··▃▂·▂▆▆█·▄ beats 1/15	medium · avg 63%	▇ 1.4%	at its level
Open Persian LLM Leaderboard v2 average (21 tasks)	50.5% ✎	▂▄▂▃█▄▃▆▃▅▂▂ beats 34/81	medium · avg 58%	▇ 1.4%	blunt · at its level
MedXPertQA MM lab	23.5% ✎	▂···▃·▃··▆▂█ beats 1/13	very hard · avg 30%	▇ 1.3%	sharp
Humanity's Last Exam	2.5% ✎	▄▄·▄··▄·▄·▄█ beats 1/7	very hard · avg 3.8%	▇ 1.2%	too hard for it
MRCR v2 8-needle, 128k lab	19.1% ✎	▄▄·▄···█···▄ beats 1/5	hard · avg 22%	▇ 1.2%	typical
Big Bench Extra Hard micro avg lab	21.9% ✎	█··▄···▄·▄·▄ beats 1/5	hard · avg 28%	▇ 1.1%	typical
SuperGPQA	30.3% ✎	▃·▃·▃▅█····▃ beats 1/8	hard · avg 32%	▇ 1.0%	at its level
OJBench	11.6% ✎	▃····▃▅··█▃▃ beats 3/8	very hard · avg 12%	▇ 0.9%	typical
Claw-Eval average score	53.1% ✎	▄·▄··█▄▄▄▄▄▄ beats 0/9	medium · avg 53%	▇ 0.7%	blunt · at its level
LongBench v2	33.2% ✎	▃······▅▅▃·█ beats 4/8	very hard · avg 32%	▇ 0.6%	too hard for it

$ where to run it

Through an API

api · Amazon Bedrock · $0.050/M · $0.04 in / $0.08 out per 1M tokens · 131k context · docs.aws.amazon.com ↗ · price sourceUSD per 1M tokens, Amazon Bedrock on-demand Gemma 4 E2B, US East (N. Virginia)/Ohio and US West (Oregon); Frankfurt and other regions priced higher. Currency: USD. Amazon Bedrock official pricing page, observed 2026-09-25.

llama.cpp · recommended

version

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 ↗
Run Gemma 4 E2B · needs ~3.4 GB · from huggingface.co/lmstudio-community
chat in the terminal
llama cli -hf lmstudio-community/gemma-4-E2B-it-GGUF:Q4_K_M
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf lmstudio-community/gemma-4-E2B-it-GGUF:Q4_K_M

LM Studio

version

Install LM Studio · Open the app once before using the lms command.
download it (macOS / Windows / Linux) ↗
Run Gemma 4 E2B · needs ~3.4 GB · from huggingface.co/lmstudio-community
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/lmstudio-community/gemma-4-E2B-it-GGUF@Q4_K_M -y
Then run lms ls to see its name, and lms chat followed by that name

Ollama

version

Install Ollama
download it (macOS / Windows) ↗
Linux
curl -fsSL https://ollama.com/install.sh | sh
Run Gemma 4 E2B · needs ~3.56 GB · from ollama.com
chat in the terminal
ollama run gemma4:e2b

MLX (Mac)

version

Install MLX (Mac) · Apple silicon Macs only, text only.
with uv
uv tool install mlx-lm
or
pip install mlx-lm
Run Gemma 4 E2B · needs ~4.3 GB, Mac only · from huggingface.co/lmstudio-community
chat in the terminal
mlx_lm.chat --model lmstudio-community/gemma-4-E2B-it-MLX-4bit
Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model lmstudio-community/gemma-4-E2B-it-MLX-4bit

Official weights and documentation: huggingface.co/google ↗

$ how the score was worked out · Gemma Team — Gemma 4 Technica… 30%, openbmb/MiniCPM5-2B model card 29%, XHToken/Spark-X2.5-4B model c… 14% and 9 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

source	results	share of the score	licence
Gemma Team — Gemma 4 Technical Report · lab-reported	19	▇▇▇▇▇▇▇▇▇ 30%	CC BY 4.0
openbmb/MiniCPM5-2B model card	20	▇▇▇▇▇▇▇▇▇ 29%	Apache-2.0
XHToken/Spark-X2.5-4B model card	11	▇▇▇▇ 14%	Apache-2.0
CohereLabs/North-Micro-Vision-Instruct model card (benchmark table)	13	▇▇▇ 10%	Apache-2.0
LiquidAI/LFM2.5-2.6B model card	7	▇▇ 7%	LFM Open License v1.0
EuroEval leaderboards (results collection), English generative leaderboard	2	▇ 5%	MIT
BRIDGE Medical Leaderboard (Harvard Medical School / Brigham and Women's YLab)	1	▇ 2%	MIT
google/gemma-4-31B-it model card · lab-reported	1	▇ 1%	Apache-2.0
Open Persian LLM Leaderboard v2 (Part DP AI and AUT NLP Lab)	1	▇ 1%	Apache-2.0
ContractBench: Can LLM Agents Preserve Observation Contracts? (arXiv:2605.17281v2)	1	▇ 1%	CC BY 4.0
LiquidAI/LFM2.5-8B-A1B model card	1	▇ 0%	LFM Open License v1.0
T1-Bench: Benchmarking Multi-Scenario Agents in Real-World Domains (arXiv:2606.11070v1)	1	▇ 0%	CC BY 4.0
prior · pull toward the middle		▇ 0%
weighted average		= 46.7

$ improve this page

Update model information Add a place to run it Add a test result Propose a benchmark Correct a benchmark score Edit 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 2026-10-11 02:20 UTC · sources and licences · json