whatmodel.sh / decision / garas (Gemma 4 31B)

garas (Gemma 4 31B)

+ contribute

garas (Gemma 4 31B) is an open-weights decision model from plantwaterer. Its published size is 31.27B parameters. Licence: apache-2.0.

Catalogue identity notes

Exact Decision Index tested engine garas; reported model garas (Gemma 4 31B). Type: LoRA; technique: autoregressive. Comparison entry; no current run command, API price or speed inferred from the benchmark. Current repository metadata checked separately; an unpinned alias is not evidence that today’s weights reproduce this historical run.

Model information sources

· Weights licence ↗ · Parameter count ↗

Publisher parameter count; includes the served backbone, not just adapter size.

60.8±1.3#102 of 355 decision modelsfrom 42 benchmark results across 1 source

$ how it compares

The decision models ranked either side of garas (Gemma 4 31B), out of 357 with a score, then more from plantwaterer.

#	model	score	where	RAM	released
100	Surogate Rune 26B-A4B v3 Surogate	61±1	request access
101	Verdin-4B v1.1 victorbennett	61±1	model files
102	garas (Gemma 4 31B) plantwaterer	61±1	model files
103	vLLM-SR d3-flash (9B) vLLM Semantic Router	61±1	local	~18GB
104	glm-5.3-flash (DecisionBench openrout) Z.ai	61±2	comparison only		2026-08-26
	more from plantwaterer
36	oenogarum (Gemma 4 31B) plantwaterer	67±1	model files
48	allec-oavg (Gemma 4 31B) plantwaterer	66±1	model files
87	garum (Gemma 4 31B) plantwaterer	62±1	model files

see all decision models · find the ones that fit your machine

$ what it's good at

tools & automation▁▃▁▁▁▂▃▄▆██▆ #74 of 250 · 6.7% of its score · 6 tests

Beats Sieve 27B v2, allec-oavg (Gemma 4 31B), Jev 1.13 here, though they rank higher overall. Trails baselabs joey-1 (2B), Hopper 12B (trained), Solomon v1.1 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: API-Bank (accuracy) submitted	84.5% ✎	▁▁▁▁▁▁▁▂▃▂█▂ beats 148/236	easy · avg 70%	▇ 2.0%	sharp · noisy
Decision Index 0.4.1: BFCL (case exact accuracy) submitted	96.7% ✎	▁▂▁▁▁▂▂▄▅▆█▁ beats 173/246	very easy · avg 92%	▇ 1.4%	sharp · maxed out · noisy
Decision Index 0.4.1: Home appliance simulator (case exact accuracy) submitted	64.8% ✎	█▁▁▃▄▄▆▄▁▂▁▃ beats 161/243	medium · avg 37%	▇ 1.2%	sharp · at its level · noisy
Decision Index 0.4.1: ToolRet (nDCG@10) submitted	65.8% ✎	▁▁▁▁▁·▁▁▁▄█▁ beats 157/246	medium · avg 60%	▇ 1.1%	blunt · at its level
Decision Index 0.4.1: When2Call (accuracy) submitted	68.0% ✎	▁▁▁▂▃▅▅▆█▇█▃ beats 91/246	easy · avg 72%	▇ 1.0%	noisy
Decision Index 0.4.1: RouterBench (selected quality (quality objective)) submitted	80.0% ✎	no other model yet		—	failed run: Publisher excludes this protocol because its prompt reveals GPT-4 as the best route. Evidence: https://huggingface.co/spaces/multimodalart/jev-decision-index/blob/c130faf65f82e81d3bdefb01e758c712dc5c5199/data/index.json

retrieval & classification▁▁▁·▁▁▁▁▃▃█▅ #132 of 258 · 7.1% of its score · 7 tests

Beats Quyet-1.0-Large, Tern-35B-A3B here, though they rank higher overall. Trails vLLM-SR vela3-fast (Vela 3.0, 0.4B), crystal-0.8b, Kas 4B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: BRIGHT (nDCG@10) submitted	50.1% ✎	▁▁▁▁▁▁▁▂▂▄▇█ beats 234/246	medium · avg 38%	▇ 2.8%	blunt · at its level
Decision Index 0.4.1: Amazon ESCI (macro-F1) submitted	52.3% ✎	▁▁▁▁▁▁▂▂▃▃█▅ beats 124/246	medium · avg 47%	▇ 2.5%	at its level
Decision Index 0.4.1: HoVer (accuracy) submitted	64.3% ✎	▆▆▇▅▄▄▃▇▄█▆▄ beats 75/246	easy · avg 74%	▇ 0.6%	blunt
Decision Index 0.4.1: PhishNChips phishing decisions (accuracy) submitted	85.2% ✎	█▅▅▄▃▂▁▁▁▁·▁ beats 191/245	easy · avg 70%	▇ 0.6%	noisy
Decision Index 0.4.1: CLINC150+OOS (macro-F1) submitted	93.4% ✎	▁▁▁▁▁▁▁▂▃▆█▄ beats 182/238	easy · avg 83%	▇ 0.3%	sharp · noisy
Decision Index 0.4.1: BANKING77 (macro-F1) submitted	87.5% ✎	▁·▁▁▁▁▁▁▂█▇█ beats 156/238	easy · avg 79%	▇ 0.3%	noisy
Decision Index 0.4.1: SGD/SGD-X (macro-F1) submitted	8.3% ✎	no other model yet		—	failed run: Publisher reports a builder bug exposing dialogue turns after the classified turn; awaiting corrected rerun. Evidence: https://huggingface.co/spaces/multimodalart/jev-decision-index/blob/c130faf65f82e81d3bdefb01e758c712dc5c5199/data/index.json

language understanding▂▂▂▃▂▅▆▃█▆▄▃ #126 of 256 · 22% of its score · 9 tests

Beats Shisa DE-2.1 (repeat) here, though they rank higher overall. Trails vLLM-SR d3-nano (2B), JevK5 v0.3, Decider 4B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: ANLI (macro-F1) submitted	71.5% ✎	▁▂▃▅▅█▂····▁ beats 169/246	medium · avg 60%	▇▇ 5.5%	at its level
Decision Index 0.4.1: NLI4CT (macro-F1) submitted	81.0% ✎	▁▁▁▂▂▃▃▄▇▇█▄ beats 161/246	easy · avg 74%	▇▇ 5.3%	typical
Decision Index 0.4.1: HellaSwag (accuracy) submitted	89.4% ✎	▂▄▂▂▄▇▃▇█▅▃▁ beats 128/246	very easy · avg 85%	▇▇ 4.0%	sharp · maxed out
Decision Index 0.4.1: iSarcasmEval (Sarcasm F1 · track A, English) submitted	65.8% ✎	▁·▁·▁▁▂▄▅█▇▄ beats 209/246	medium · avg 44%	▇ 2.6%	at its level
Decision Index 0.4.1: ACOS (per-review F1) submitted	22.5% ✎	▁··▂▃▃▅▇█▇▃▃ beats 128/246	hard · avg 22%	▇ 1.5%	at its level · noisy
Decision Index 0.4.1: VAST (macro-F1) submitted	67.9% ✎	▁▁▁▂▅▄▄▂▄▆█▄ beats 125/246	medium · avg 62%	▇ 1.1%	at its level
Decision Index 0.4.1: ContractNLI (macro-F1) submitted	70.2% ✎	▁▁▁▁▁▁▁▂▂▇▇█ beats 75/246	easy · avg 71%	▇ 1.0%	noisy
Decision Index 0.4.1: RAGTruth (F1 on hallucinated class) submitted	72.1% ✎	▁▁▁▁▁▁▁▁▁▃▆█ beats 110/246	easy · avg 66%	▇ 0.8%	noisy

1 more test

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: FinEntity (macro-F1) submitted	93.2% ✎	▁···▁▁▁▁▃▆█▃ beats 196/246	easy · avg 84%	▇ 0.6%	noisy

arts & human taste▁▇▄▃▅▅█▅▄▂▁▁ #72 of 247 · 10% of its score · 6 tests

Beats Jev 1.13, Quyet-1.0-Large, jev-brain · Gemma-4-31B (decision readout) here, though they rank higher overall. Trails Amber 0.4B, Crivo-1B, Kas 4B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: New Yorker (accuracy) submitted	75.2% ✎	▁▁▂▃▃▃▇▅█▅▃▁ beats 192/246	medium · avg 61%	▇▇ 4.1%	at its level
Decision Index 0.4.1: Humicroedit (accuracy) submitted	65.0% ✎	▁·▁▃▂▂▅█▄▂▁▁ beats 204/246	medium · avg 60%	▇ 2.1%	blunt · at its level
Decision Index 0.4.1: cfcolor (accuracy) submitted	65.8% ✎	▁········▁▇█ beats 199/246	medium · avg 60%	▇ 1.2%	blunt · at its level
Decision Index 0.4.1: POP909-CL (accuracy) submitted	23.6% ✎	▄▅▅▃▆▅▃▅▂▅▃█ beats 134/237	hard · avg 21%	▇ 1.2%	sharp · at its level · noisy
Decision Index 0.4.1: BPoMP (accuracy) submitted	92.8% ✎	▁█▄▄▅█▆▅▄▁▁▁ beats 175/246	easy · avg 83%	▇ 1.0%	sharp · noisy
Decision Index 0.4.1: Habermas Machine (accuracy) submitted	48.6% ✎	▁▂▄█▇▅▁▁▄▄▁▁ beats 173/246	medium · avg 47%	▇ 0.4%	blunt · at its level

knowledge & reasoning▁▂▂▂▂▃▃▅▅▆█▂ #74 of 263 · 44% of its score · 14 tests

Beats StartLux-Decision v1.1 9B, Sieve 27B v2, KnowLine 9B Gen1 here, though they rank higher overall. Trails Solar Decide, Clef-flash, Clef-omni here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: MMLU-Pro (accuracy) submitted	64.8% ✎	▄▄▃▃▂▄▅▇██▅▁ beats 177/246	medium · avg 46%	▇▇▇▇ 11%	sharp · at its level
Decision Index 0.4.1: MMLU (accuracy) submitted	84.7% ✎	▁▄▄▃▃▂▅▇▅█▇▁ beats 186/246	easy · avg 70%	▇▇▇ 6.5%	sharp
Decision Index 0.4.1: BBH (accuracy) submitted	72.7% ✎	▄▆▄▄█▇▇▄▃▁▁▁ beats 149/246	easy · avg 66%	▇▇ 6.2%	typical
Decision Index 0.4.1: GSM8K (accuracy (4- and 10-choice)) submitted	62.9% ✎	▁··▄█▆▆▆▅▅▃▁ beats 168/246	medium · avg 45%	▇ 3.7%	sharp · at its level
Decision Index 0.4.1: GPQA Diamond (accuracy) submitted	43.9% ✎	▂▆▅▅▆█▆▃▁··▁ beats 140/246	medium · avg 39%	▇ 3.6%	blunt · at its level
Decision Index 0.4.1: WinoGrande (accuracy) submitted	87.3% ✎	█▅█▅▄█▆▄▃▃▁▁ beats 176/246	easy · avg 78%	▇ 2.9%	typical
Decision Index 0.4.1: ARC-Challenge (accuracy) submitted	97.0% ✎	▂▂▂▂▂▂▁▃▄▆█▆ beats 198/246	very easy · avg 90%	▇ 2.7%	sharp · maxed out
Decision Index 0.4.1: SATA-Bench (case exact accuracy) submitted	28.8% ✎	▁▁▁▁▁▁▁▂▅█▂▁ beats 130/246	hard · avg 22%	▇ 1.7%	at its level

6 more tests

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: MuSR (accuracy) submitted	62.9% ✎	▁▁▂█▅▄▁▁▁▁▁▁ beats 158/246	medium · avg 60%	▇ 1.3%	blunt · at its level
Decision Index 0.4.1: ARC-Easy (accuracy) submitted	99.1% ✎	▁▁▂▁▂▂▂▂▃▆█▅ beats 212/246	very easy · avg 96%	▇ 1.2%	sharp · maxed out · noisy
Decision Index 0.4.1: CRUXEval (accuracy) submitted	79.3% ✎	██▄▃▃▄▁▁···▁ beats 194/246	medium · avg 58%	▇ 1.1%	at its level · noisy
Decision Index 0.4.1: CLadder (accuracy) submitted	76.3% ✎	▁███▆▅▃▃▁▁▁▁ beats 161/246	easy · avg 71%	▇ 1.0%	typical
Decision Index 0.4.1: ChessBench (accuracy) submitted	20.0% ✎	▄▇▄▄▄▆▄█▄▁▁▁ beats 170/246	hard · avg 16%	▇ 0.8%	blunt
Decision Index 0.4.1: HLE (accuracy) submitted	11.2% ✎	▁▁▂▃▆█▇▅▅▃▁▁ beats 70/246	very hard · avg 13%	▇ 0.0%	blunt

other tests2 not in a task · 9.3% of its score

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: private same skills (equated points)	56.2400 points ✎	▂▂▂▃▃▃▄▅▅▇█▃ beats 182/247	avg 41.206 points	▇▇ 6.0%	blunt
Decision Index 0.4.1: private new domains (equated points)	63.0300 points ✎	▃█▃▃▂▆▃▃▄▄▄▄ beats 196/247	avg 38.861 points	▇ 3.3%	blunt

$ where to run it

Download model files ↗
Published model files verified. Includes adapters or readout weights; follow the model documentation for the required base model. This is a verified repository snapshot; the benchmark does not establish an identical weight revision. A local run command and hardware requirements have not been verified for this entry. Source · checked 2026-10-31.

Model repository and documentation: huggingface.co/plantwaterer ↗

$ how the score was worked out · Decision Index 0.4.1 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
Decision Index 0.4.1 · submitted; independent execution unverified	42	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 99%	CC BY 4.0
prior · pull toward the middle		▇ 1%
weighted average		= 60.8

not counted (2)

Decision Index 0.4.1	Decision Index 0.4.1: RouterBench (selected quality (quality objective)) [correct]	80.0%	failed run: Publisher excludes this protocol because its prompt reveals GPT-4 as the best route. Evidence: https://huggingface.co/spaces/multimodalart/jev-decision-index/blob/c130faf65f82e81d3bdefb01e758c712dc5c5199/data/index.json
Decision Index 0.4.1	Decision Index 0.4.1: SGD/SGD-X (macro-F1) [correct]	8.3%	failed run: Publisher reports a builder bug exposing dialogue turns after the classified turn; awaiting corrected rerun. Evidence: https://huggingface.co/spaces/multimodalart/jev-decision-index/blob/c130faf65f82e81d3bdefb01e758c712dc5c5199/data/index.json

$ 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 2026-11-03 02:37 UTC · sources and licences · json