whatmodel.sh / DiffusionGemma 26B-A4B

DiffusionGemma 26B-A4B

+ contribute

DiffusionGemma 26B-A4B is an open-weights language model from Google, released June 2026. Its published size is 25.82B parameters (3.8B active per token) and a 262k-token context window. Licence: apache-2.0.

Catalogue identity notes

google/diffusiongemma-26B-A4B-it @f7f5b7f (created 2026-06-09, licence apache-2.0). Google's block-diffusion Gemma 4 variant (June 2026; 25.2B total / 3.8B active per the card; text and image input). Comparison-only for now: the card's inference path is transformers' DiffusionGemmaForBlockDiffusion (no one-line command) and the GGUF needs a llama.cpp branch with llama-diffusion-cli, so no run option is claimed. Lab rows from its own card, paired with the catalogued Gemma 4 26B-A4B column.

Model information sources

Release date ↗ · Weights licence ↗

Card: 'Total Parameters 25.2B', 'Active Parameters 3.8B' (HF safetensors metadata 25,823,778,864 incl. the ~550M vision encoder).

58.5±1.4comparison only: no current run option verifiedfrom 12 benchmark results across 1 source

$ how it compares

The LLMs ranked either side of DiffusionGemma 26B-A4B, out of 828 with a score, then more from Google.

#	model	score	where	RAM	released
254	Qwen3 14B Alibaba	59±1	api $0.61/M +local	~10GB	2025-04
255	Mercury 2 Inception	59±2	comparison only
256	DiffusionGemma 26B-A4B Google	59±1	comparison only		2026-06
257	gpt-oss-20b OpenAI	58±1	api $0.13/M +local	~14GB	2025-08
258	Llama 3.3 Nemotron Super 49B v1 NVIDIA	58±2	local	~22GB	2025-03
	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▁▁▂▆█▆▅▄▄▃▂▁ #190 of 340 * · 10% of its score · 1 test

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.

test	result	vs the others in its run	difficulty	counts	why
τ²-bench aggregate lab	56.2% ✎	▁▁▁·▂▂▄▃▃▅█▂ beats 7/37	medium · avg 36%	▇▇▇▇ 10%	at its level · small run

coding▁▁▁▃▅▇█▇▅▄▃▁ #211 of 524 * · 9.2% of its score · 1 test

Beats Gemini 2.5 Flash, Qwen3.5 4B, Qwen3 30B-A3B Instruct 2507 here, though they rank higher overall. Trails QwQ-32B-Preview, OpenThinker 32B, Command A Reasoning (08-2025) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LiveCodeBench v6 (window unspecified) lab	69.1% ✎	▂▃▄▄▅██▅▄▇▃▆ beats 33/65	medium · avg 47%	▇▇▇▇ 9.2%	at its level · small run

math▁▂▃▅▇▅▅█▄▁▁▁ #209 of 462 * · 19% of its score · 2 tests

Beats MiniMax-M2.1, DeepSeek V3 0324, Qwen3 235B-A22B (Apr 2025, hybrid) here, though they rank higher overall. Trails Phi-4-mini-reasoning 3.8B, OpenThinker3 7B, Olmo 3 7B Think here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
AIME 2026 lab	69.1% ✎	▁▁▁▂▃▂▆█▆▂·▁ beats 8/49	medium · avg 51%	▇▇▇▇▇ 12%	sharp · small run
MATH-Vision lab	70.5% ✎	▁▁·▁··▄▁▃▇▇█ beats 8/24	medium · avg 52%	▇▇▇ 7.2%	at its level · small run

vision & documents▁▁▃▃▂▃▅▄▇▇█▃ #117 of 182 · 15% of its score · 3 tests

Beats DeepSeek V4 Flash Vision (Exp) here, though they rank higher overall. Trails Llama 4 Maverick (402B-A17B), GPT-4o (2024-11-20), Qwen3-VL 8B Thinking here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
MATH-Vision lab	70.5% ✎	▁▁·▁··▄▁▃▇▇█ beats 8/24	medium · avg 52%	▇▇▇ 7.2%	at its level · small run
MMMU Pro lab	54.3% ✎	▂▃▁▃▆▇▃▇▇██▅ beats 17/52	medium · avg 46%	▇▇ 5.0%	at its level · small run
OmniDocBench 1.5 ↓ lab	31.9% ✎	███·····████ beats 1/6	avg 35%	▇ 2.5%	blunt · small run

other tests6 not in a task · 54% of its score

test	result	vs the others in its run	difficulty	counts	why
GPQA Diamond lab	73.2% ✎	▁▁▁▁▂▄▅▅███▄ beats 165/310	hard · avg 49%	▇▇▇▇▇▇ 16%	sharp · at its level · small run
MedXPertQA MM lab	49.0% ✎	▂···▂·▃··▆▂█ beats 6/13	hard · avg 32%	▇▇▇▇ 11%	sharp · at its level · small run
MMLU-Pro lab	77.6% ✎	▁▁·▁▁▁▃▂▄▅█▁ beats 111/188	medium · avg 60%	▇▇▇ 7.9%	at its level · small run
MRCR v2 8-needle, 128k lab	32.0% ✎	█▄▄█·▄···▄·▄ beats 3/8	hard · avg 22%	▇▇▇ 7.1%	at its level · small run
MMMLU lab	81.5% ✎	▂▁▁▂▃▂▄▄▅██▃ beats 33/46	medium · avg 67%	▇▇▇ 6.8%	at its level · small run
Humanity's Last Exam lab	11.0% ✎	▁▁▁▃▆▆▇██▆▁▃ beats 20/104	very hard · avg 4.1%	▇▇ 5.7%	small run

$ where to run it

No verified API or download recorded yet · add one

Official weights and documentation: huggingface.co/google ↗

$ how the score was worked out · google/diffusiongemma-26B-A4B… 99%

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
google/diffusiongemma-26B-A4B-it model card · lab-reported	12	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 99%	Apache-2.0
prior · pull toward the middle		▇ 1%
weighted average		= 58.5

$ 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