whatmodel.sh / decision / decisio 0.4.0 (Gemma 4 12B) (decision readout)

decisio 0.4.0 (Gemma 4 12B) (decision readout)

+ contribute

decisio 0.4.0 (Gemma 4 12B) (decision readout) is a source-available decision implementation from aminry, released 4 October 2026. Its published size is 11.96B parameters.

Catalogue identity notes

Exact Decision Index tested engine decisio-0.4.0-gemma-4-12b; reported model decisio 0.4.0 (Gemma 4 12B). Type: inference technique; technique: autoregressive. Implementation: https://github.com/aminry/decisio. Comparison entry; no current run command, API price or speed inferred from the benchmark.

Model information sources

Release date ↗ · Parameter count ↗

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

54.4±1.2#164 of 355 decision modelsfrom 42 benchmark results across 1 source

$ how it compares

The decision models ranked either side of decisio 0.4.0 (Gemma 4 12B) (decision readout), out of 357 with a score, then more from aminry.

#	model	score	where	released
164	ezjev 4B s2 everettjf	55±1	model files
165	JoshuaSP diffgemma (decision readout) JoshuaSP	55±1	source code
166	decisio 0.4.0 (Gemma 4 12B) (decision readout) aminry	54±1	source code	2026-10-04
167	JevK5 9B v0.3.3 alibiserikbay	54±1	model files
168	Jevfire Qwen3.8-27B (decision readout) kikoncuo	54±1	source code
	more from aminry
81	decisio 0.6.0 (Gemma 4 31B) (decision readout) aminry	63±1	source code	2026-10-05
149	decisio 0.4.0 (Qwen3.6-35B-A3B) (decision readout) aminry	56±1	source code	2026-10-04

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

$ what it's good at

tools & automation▁▃▁▁▁▂▃▄▆██▆ #94 of 250 · 7.5% of its score · 6 tests

Beats Jev 1.13, Xor 26B-A4B v1.2, Senova-4B v2 here, though they rank higher overall. Trails OneJev1.6-0.8B, Jev-Omni, baselabs joey-1 (2B) 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	87.0% ✎	▁▁▁▁▁▁▁▂▃▂█▂ beats 207/236	easy · avg 70%	▇ 2.3%	sharp · noisy
Decision Index 0.4.1: BFCL (case exact accuracy) submitted	96.2% ✎	▁▂▁▁▁▂▂▄▅▆█▁ beats 154/246	very easy · avg 92%	▇ 2.0%	sharp · maxed out · noisy
Decision Index 0.4.1: Home appliance simulator (case exact accuracy) submitted	44.3% ✎	█▁▁▃▄▄▆▄▁▂▁▃ beats 120/243	medium · avg 37%	▇ 1.2%	sharp · at its level · noisy
Decision Index 0.4.1: When2Call (accuracy) submitted	76.1% ✎	▁▁▁▂▃▅▅▆█▇█▃ beats 119/246	easy · avg 72%	▇ 1.1%	noisy
Decision Index 0.4.1: ToolRet (nDCG@10) submitted	62.2% ✎	▁▁▁▁▁·▁▁▁▄█▁ beats 91/246	medium · avg 60%	▇ 1.0%	blunt · at its level
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▁▁▁·▁▁▁▁▃▃█▅ #116 of 258 · 6.4% of its score · 7 tests

Beats AutoTrust-JEV-27B v1.1, Quyet-1.0-Large, deck31b (Gemma 4 31B) (decision readout) 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	47.1% ✎	▁▁▁▁▁▁▁▂▂▄▇█ beats 184/246	medium · avg 38%	▇ 2.4%	blunt · at its level
Decision Index 0.4.1: Amazon ESCI (macro-F1) submitted	53.5% ✎	▁▁▁▁▁▁▂▂▃▃█▅ beats 139/246	medium · avg 47%	▇ 2.2%	at its level
Decision Index 0.4.1: HoVer (accuracy) submitted	69.3% ✎	▆▆▇▅▄▄▃▇▄█▆▄ beats 96/246	easy · avg 74%	▇ 0.6%	blunt
Decision Index 0.4.1: PhishNChips phishing decisions (accuracy) submitted	81.3% ✎	█▅▅▄▃▂▁▁▁▁·▁ beats 177/245	easy · avg 70%	▇ 0.5%	at its level · noisy
Decision Index 0.4.1: CLINC150+OOS (macro-F1) submitted	87.2% ✎	▁▁▁▁▁▁▁▂▃▆█▄ beats 93/238	easy · avg 83%	▇ 0.4%	sharp · noisy
Decision Index 0.4.1: BANKING77 (macro-F1) submitted	72.9% ✎	▁·▁▁▁▁▁▁▂█▇█ beats 54/238	easy · avg 79%	▇ 0.3%	noisy
Decision Index 0.4.1: SGD/SGD-X (macro-F1) submitted	4.7% ✎	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▂▂▂▃▂▅▆▃█▆▄▃ #135 of 256 · 23% of its score · 9 tests

Beats Solar Decide, Winnow-12B, JoshuaSP diffgemma (decision readout) here, though they rank higher overall. Trails Sieve 2B v3, vLLM-SR d3-nano (2B), JevK5 v0.3 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: NLI4CT (macro-F1) submitted	81.3% ✎	▁▁▁▂▂▃▃▄▇▇█▄ beats 163/246	easy · avg 74%	▇▇ 5.4%	typical
Decision Index 0.4.1: HellaSwag (accuracy) submitted	86.7% ✎	▂▄▂▂▄▇▃▇█▅▃▁ beats 117/246	very easy · avg 85%	▇▇ 5.4%	sharp
Decision Index 0.4.1: ANLI (macro-F1) submitted	70.2% ✎	▁▂▃▅▅█▂····▁ beats 160/246	medium · avg 60%	▇▇ 5.3%	at its level
Decision Index 0.4.1: iSarcasmEval (Sarcasm F1 · track A, English) submitted	42.7% ✎	▁·▁·▁▁▂▄▅█▇▄ beats 102/246	medium · avg 44%	▇ 2.3%	at its level
Decision Index 0.4.1: ACOS (per-review F1) submitted	15.5% ✎	▁··▂▃▃▅▇█▇▃▃ beats 82/246	hard · avg 22%	▇ 1.1%	at its level · noisy
Decision Index 0.4.1: ContractNLI (macro-F1) submitted	74.3% ✎	▁▁▁▁▁▁▁▂▂▇▇█ beats 107/246	easy · avg 71%	▇ 1.1%	at its level · noisy
Decision Index 0.4.1: VAST (macro-F1) submitted	59.1% ✎	▁▁▁▂▅▄▄▃▄▆█▄ beats 103/246	medium · avg 62%	▇ 1.0%	at its level
Decision Index 0.4.1: RAGTruth (F1 on hallucinated class) submitted	69.6% ✎	▁▁▁▁▁▁▁▁▁▃▆█ beats 94/246	easy · avg 66%	▇ 0.8%	at its level · noisy

1 more test

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: FinEntity (macro-F1) submitted	90.5% ✎	▁···▁▁▁▁▃▆█▃ beats 153/246	easy · avg 84%	▇ 0.7%	noisy

arts & human taste▁▇▄▃▅▅█▅▄▂▁▁ #98 of 247 · 9.3% of its score · 6 tests

Beats Jev 1.13, Jeffan-4B-2010, Bobcat 1.3 (31B) here, though they rank higher overall. Trails vLLM-SR vela3-fast-vision (Vela 3.0, 0.25B), Crivo-400M, Amber 0.4B 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	63.3% ✎	▁▁▂▃▃▃▇▅█▅▃▁ beats 121/246	medium · avg 61%	▇▇ 4.0%	at its level
Decision Index 0.4.1: Humicroedit (accuracy) submitted	61.7% ✎	▁·▁▃▂▂▅█▄▂▁▁ beats 140/246	medium · avg 60%	▇ 1.9%	blunt · at its level
Decision Index 0.4.1: BPoMP (accuracy) submitted	90.7% ✎	▁█▄▄▅█▆▅▄▁▁▁ beats 150/246	easy · avg 83%	▇ 1.2%	sharp · noisy
Decision Index 0.4.1: cfcolor (accuracy) submitted	63.0% ✎	▁········▁▇█ beats 143/246	medium · avg 60%	▇ 1.1%	blunt · at its level
Decision Index 0.4.1: POP909-CL (accuracy) submitted	6.2% ✎	▄▅▅▃▆▅▃▅▂▅▃█ beats 81/237	hard · avg 21%	▇ 0.8%	sharp · at its level · noisy
Decision Index 0.4.1: Habermas Machine (accuracy) submitted	46.5% ✎	▁▂▄█▇▅▁▁▄▄▁▁ beats 152/246	medium · avg 47%	▇ 0.3%	blunt · at its level

knowledge & reasoning▁▂▂▂▂▃▃▅▅▆█▂ #119 of 263 · 45% of its score · 14 tests

Beats OneJev1.6-4B, Winnow-12B (DecisionBench b6ac22b0), Cerno 4B v1.1 here, though they rank higher overall. Trails PurpleMIST-Flash-1.0, Jevfire Qwen3.8-27B (decision readout), JevK5 9B v0.3.3 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	55.0% ✎	▄▄▃▃▂▄▅▇██▅▁ beats 135/246	medium · avg 46%	▇▇▇▇ 10%	sharp · at its level
Decision Index 0.4.1: MMLU (accuracy) submitted	78.3% ✎	▁▄▄▃▃▂▅▇▅█▇▁ beats 142/246	easy · avg 70%	▇▇▇ 7.1%	sharp
Decision Index 0.4.1: BBH (accuracy) submitted	69.1% ✎	▄▆▄▄█▇▇▄▃▁▁▁ beats 127/246	easy · avg 66%	▇▇▇ 6.4%	at its level
Decision Index 0.4.1: ARC-Challenge (accuracy) submitted	95.0% ✎	▂▂▂▂▂▂▁▃▄▆█▆ beats 140/246	very easy · avg 90%	▇▇ 4.1%	sharp · maxed out
Decision Index 0.4.1: GSM8K (accuracy (4- and 10-choice)) submitted	38.7% ✎	▁··▄█▆▆▆▅▅▃▁ beats 103/246	medium · avg 45%	▇ 3.4%	sharp · at its level
Decision Index 0.4.1: WinoGrande (accuracy) submitted	73.0% ✎	█▅█▅▄█▆▄▃▃▁▁ beats 100/246	easy · avg 78%	▇ 3.2%	typical
Decision Index 0.4.1: GPQA Diamond (accuracy) submitted	38.8% ✎	▂▆▅▅▆█▆▃▁··▁ beats 105/246	medium · avg 39%	▇ 3.1%	blunt · at its level
Decision Index 0.4.1: ARC-Easy (accuracy) submitted	97.8% ✎	▁▁▂▁▂▂▂▂▃▆█▅ beats 116/246	very easy · avg 96%	▇ 2.1%	sharp · maxed out · noisy

6 more tests

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: SATA-Bench (case exact accuracy) submitted	26.4% ✎	▁▁▁▁▁▁▁▂▅█▂▁ beats 112/246	hard · avg 22%	▇ 1.3%	at its level
Decision Index 0.4.1: MuSR (accuracy) submitted	58.1% ✎	▁▁▂█▅▄▁▁▁▁▁▁ beats 107/246	medium · avg 60%	▇ 1.2%	blunt · at its level
Decision Index 0.4.1: CRUXEval (accuracy) submitted	65.4% ✎	██▄▃▃▄▁▁···▁ beats 156/246	medium · avg 58%	▇ 1.1%	at its level · noisy
Decision Index 0.4.1: CLadder (accuracy) submitted	63.4% ✎	▁███▆▅▃▃▁▁▁▁ beats 83/246	easy · avg 71%	▇ 1.0%	at its level
Decision Index 0.4.1: ChessBench (accuracy) submitted	16.5% ✎	▄▇▄▄▄▆▄█▄▁▁▁ beats 125/246	hard · avg 16%	▇ 0.6%	blunt
Decision Index 0.4.1: HLE (accuracy) submitted	10.6% ✎	▁▁▂▃▆█▇▅▅▃▁▁ beats 48/246	very hard · avg 13%	▇ 0.0%	blunt

other tests2 not in a task · 8.1% of its score

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: private same skills (equated points)	47.9400 points ✎	▂▂▂▃▃▃▄▅▅▇█▃ beats 136/247	avg 41.206 points	▇▇ 5.3%	blunt
Decision Index 0.4.1: private new domains (equated points)	49.9500 points ✎	▃█▃▃▂▆▃▃▄▄▄▄ beats 159/247	avg 38.861 points	▇ 2.9%	blunt

$ where to run it

Implementation and setup instructions ↗
Published implementation and setup documentation verified. This entry describes a decision readout or evaluated configuration, not necessarily separate trained weights. Follow the referenced version and model settings; hardware compatibility has not been verified. Source · checked 2026-10-31.

Model repository and documentation: github.com/aminry ↗

$ 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		= 54.4

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]	4.7%	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