whatmodel.sh / decision / system-one-gemma

system-one-gemma

+ contribute

system-one-gemma is a source-available decision implementation from akash-kamat. Its published size is 0.27B parameters.

Catalogue identity notes

Exact Decision Index tested engine akash-gemma; reported model system-one-gemma. Type: LoRA + head; technique: autoregressive. Implementation: https://github.com/akash-kamat/system-one-gemma. Comparison entry; no current run command, API price or speed inferred from the benchmark.

Model information sources

· Parameter count ↗

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

18.4±1.1#347 of 355 decision modelsfrom 42 benchmark results across 1 source

$ how it compares

The decision models ranked either side of system-one-gemma, out of 357 with a score.

#	model	score	where	released
347	Bekko System One v0 17M hotchpotch	19±1	model files
348	bge-reranker-v2-m3 (DecisionBench 953dc6f6) BAAI	19±2	model files
349	system-one-gemma akash-kamat	18±1	source code
350	Supersonic Labs Julia 1 Supersonic Labs	18±1	model files	2026-09-25
351	KaLM-Reranker-V1-Small-R2 (DecisionBench d1c8fdf9) KaLM-Embedding	18±2	model files

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

$ what it's good at

tools & automation▁▃▁▁▁▂▃▄▆██▆ #247 of 250 · 10% of its score · 6 tests

Beats jeff GLiFormer-large (decision readout), Bekko System One v0 17M here, though they rank higher overall. Trails LFM2.5-350M-RLCD, Nirnaya 0.4B, Supersonic Labs Julia 1 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: BFCL (case exact accuracy) submitted	34.8% ✎	▁▃▁▁▁▂▂▄▅▆█▁ beats 3/246	very easy · avg 92%	▇▇▇ 7.1%	sharp · at its level · noisy
Decision Index 0.4.1: API-Bank (accuracy) submitted	1.2% ✎	▁▃▁▁▁▁▁▂▃▂█▂ beats 2/236	easy · avg 70%	▇ 1.5%	sharp · noisy
Decision Index 0.4.1: When2Call (accuracy) submitted	15.7% ✎	▃▁▁▂▃▅▅▆█▇█▃ beats 3/246	easy · avg 72%	▇ 1.0%	at its level · noisy
Decision Index 0.4.1: ToolRet (nDCG@10) submitted	16.4% ✎	▁▁▃▁▁·▁▁▁▄█▁ beats 8/246	medium · avg 60%	▇ 0.8%	blunt · at its level
Decision Index 0.4.1: Home appliance simulator (case exact accuracy) submitted	0.0% ✎	█▁▁▃▄▄▆▄▁▂▁▃ beats 23/243	medium · avg 37%	▇ 0.1%	sharp · too hard for it · noisy
Decision Index 0.4.1: RouterBench (selected quality (quality objective)) submitted	35.8% ✎	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▁▁▁·▃▁▁▁▃▃█▅ #250 of 258 · 4.6% of its score · 7 tests

Beats DM-JEPA 1.1, JevK5 Lite, Qwen-2.5-1B-RLCD (decision readout) here, though they rank higher overall. Trails Nirnaya 0.4B, Supersonic Labs Julia 1 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: Amazon ESCI (macro-F1) submitted	19.5% ✎	▁▁▁▁▁▃▂▂▃▃█▅ beats 8/246	medium · avg 47%	▇ 1.2%	typical
Decision Index 0.4.1: BRIGHT (nDCG@10) submitted	10.3% ✎	▁▃▁▁▁▁▁▂▂▄▇█ beats 4/246	medium · avg 38%	▇ 1.2%	blunt
Decision Index 0.4.1: CLINC150+OOS (macro-F1) submitted	24.9% ✎	▁▁▁▁▁▃▁▂▃▆█▄ beats 17/238	easy · avg 83%	▇ 0.6%	sharp · at its level · noisy
Decision Index 0.4.1: HoVer (accuracy) submitted	50.5% ✎	▆▆▇▅▄▄▃▇▄█▆▄ beats 7/246	easy · avg 74%	▇ 0.6%	blunt · at its level
Decision Index 0.4.1: PhishNChips phishing decisions (accuracy) submitted	60.0% ✎	█▅▅▄▃▂▁▁▁▁·▁ beats 84/245	easy · avg 70%	▇ 0.5%	at its level · noisy
Decision Index 0.4.1: BANKING77 (macro-F1) submitted	53.5% ✎	▁·▁▁▁▁▁▁▃█▇█ beats 17/238	easy · avg 79%	▇ 0.4%	at its level · noisy
Decision Index 0.4.1: SGD/SGD-X (macro-F1) submitted	45.2% ✎	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▃▂▂▃▂▅▆▃█▆▄▃ #252 of 256 · 23% of its score · 9 tests

Beats Prumo (rule-based engine) here, though they rank higher overall. Trails Supersonic Labs Julia 1 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: HellaSwag (accuracy) submitted	28.4% ✎	▃▄▂▂▄▇▃▇█▅▃▁ beats 4/246	very easy · avg 85%	▇▇▇▇ 8.8%	sharp · at its level
Decision Index 0.4.1: NLI4CT (macro-F1) submitted	42.5% ✎	▁▁▃▂▂▃▃▄▇▇█▄ beats 6/246	easy · avg 74%	▇▇ 5.8%	at its level
Decision Index 0.4.1: ANLI (macro-F1) submitted	23.9% ✎	▁▃▃▅▅█▂····▁ beats 5/246	medium · avg 60%	▇ 3.5%	at its level
Decision Index 0.4.1: FinEntity (macro-F1) submitted	13.0% ✎	▁···▃▁▁▁▃▆█▃ beats 1/246	easy · avg 84%	▇ 1.1%	at its level · noisy
Decision Index 0.4.1: ContractNLI (macro-F1) submitted	20.7% ✎	▁▁▁▃▁▁▁▂▂▇▇█ beats 6/246	easy · avg 71%	▇ 1.0%	at its level · noisy
Decision Index 0.4.1: iSarcasmEval (Sarcasm F1 · track A, English) submitted	25.1% ✎	▁·▁·▁▁▂▄▅█▇▄ beats 36/246	medium · avg 44%	▇ 0.9%	typical
Decision Index 0.4.1: VAST (macro-F1) submitted	34.9% ✎	▁▁▁▃▅▄▄▂▄▆█▄ beats 14/246	medium · avg 62%	▇ 0.7%	at its level
Decision Index 0.4.1: RAGTruth (F1 on hallucinated class) submitted	12.9% ✎	▁▁▁▁▁▃▁▁▁▃▆█ beats 14/246	easy · avg 66%	▇ 0.6%	at its level · noisy

1 more test

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: ACOS (per-review F1) submitted	3.1% ✎	▁··▃▃▃▅▇█▇▃▃ beats 6/246	hard · avg 22%	▇ 0.2%	too hard for it · noisy

arts & human taste▁▇▄▃▅▅█▅▄▂▁▁ #201 of 247 · 7.2% of its score · 6 tests

Beats vLLM-SR d3-nano (2B), Valen-4B, Saracura PT-BR 4B here, though they rank higher overall. Trails LFM2.5-350M-RLCD 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	29.2% ✎	▁▁▃▃▃▃▇▅█▅▃▁ beats 16/246	medium · avg 61%	▇ 2.6%	at its level
Decision Index 0.4.1: BPoMP (accuracy) submitted	57.8% ✎	▁█▄▄▅█▆▅▄▁▁▁ beats 46/246	easy · avg 83%	▇ 2.0%	sharp · at its level · noisy
Decision Index 0.4.1: Humicroedit (accuracy) submitted	47.3% ✎	▁·▃▃▂▂▅█▄▂▁▁ beats 7/246	medium · avg 60%	▇ 1.5%	blunt · at its level
Decision Index 0.4.1: cfcolor (accuracy) submitted	52.4% ✎	▁········▁▇█ beats 40/246	medium · avg 60%	▇ 0.9%	blunt · at its level
Decision Index 0.4.1: Habermas Machine (accuracy) submitted	29.6% ✎	▃▂▄█▇▅▁▁▄▄▁▁ beats 2/246	medium · avg 47%	▇ 0.2%	blunt · at its level
Decision Index 0.4.1: POP909-CL (accuracy) submitted	0.3% ✎	▄▅▅▃▆▅▃▅▂▅▃█ beats 8/237	hard · avg 21%	▇ 0.0%	sharp · too hard for it · noisy

knowledge & reasoning▁▃▂▂▂▃▃▅▅▆█▂ #256 of 263 · 48% of its score · 14 tests

Beats Crivo-17M, Prumo (rule-based engine), Bekko System One v0 17M here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: ARC-Easy (accuracy) submitted	44.6% ✎	▁▃▂▁▂▂▂▂▃▆█▅ beats 7/246	very easy · avg 96%	▇▇▇▇▇▇ 16%	sharp · at its level · noisy
Decision Index 0.4.1: ARC-Challenge (accuracy) submitted	31.0% ✎	▃▂▂▂▂▂▁▃▄▆█▆ beats 9/246	very easy · avg 90%	▇▇▇▇ 10%	sharp · at its level
Decision Index 0.4.1: MMLU (accuracy) submitted	30.8% ✎	▁▄▄▃▃▂▅▇▅█▇▁ beats 10/246	easy · avg 70%	▇▇ 5.7%	sharp · at its level
Decision Index 0.4.1: BBH (accuracy) submitted	28.5% ✎	▄▆▄▄█▇▇▄▃▁▁▁ beats 0/246	easy · avg 66%	▇▇ 4.6%	at its level
Decision Index 0.4.1: WinoGrande (accuracy) submitted	49.9% ✎	█▅█▅▄█▆▄▃▃▁▁ beats 3/246	easy · avg 78%	▇▇ 3.9%	at its level
Decision Index 0.4.1: MMLU-Pro (accuracy) submitted	13.2% ✎	▄▄▃▃▂▄▅▇██▅▁ beats 8/246	medium · avg 46%	▇ 3.2%	sharp
Decision Index 0.4.1: GPQA Diamond (accuracy) submitted	24.5% ✎	▂▆▅▅▆█▆▃▁··▁ beats 14/246	medium · avg 39%	▇ 1.6%	blunt
Decision Index 0.4.1: MuSR (accuracy) submitted	41.6% ✎	▁▃▂█▅▄▁▁▁▁▁▁ beats 14/246	medium · avg 60%	▇ 1.0%	blunt · at its level

6 more tests

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: CLadder (accuracy) submitted	50.2% ✎	▁███▆▅▃▃▁▁▁▁ beats 13/246	easy · avg 71%	▇ 0.9%	at its level
Decision Index 0.4.1: GSM8K (accuracy (4- and 10-choice)) submitted	20.5% ✎	▁··▄█▆▆▆▅▅▃▁ beats 40/246	medium · avg 45%	▇ 0.9%	sharp
Decision Index 0.4.1: CRUXEval (accuracy) submitted	34.0% ✎	██▄▃▃▄▁▁···▁ beats 22/246	medium · avg 58%	▇ 0.7%	at its level · noisy
Decision Index 0.4.1: SATA-Bench (case exact accuracy) submitted	0.1% ✎	▃▁▁▁▁▁▁▂▅█▂▁ beats 9/246	hard · avg 22%	▇ 0.3%	too hard for it
Decision Index 0.4.1: ChessBench (accuracy) submitted	9.1% ✎	▄▇▄▄▄▆▄█▄▁▁▁ beats 47/246	hard · avg 16%	▇ 0.2%	blunt · too hard for it
Decision Index 0.4.1: HLE (accuracy) submitted	12.6% ✎	▁▁▂▃▆█▇▅▅▃▁▁ beats 132/246	very hard · avg 13%	▇ 0.0%	blunt

other tests2 not in a task · 6.2% of its score

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: private same skills (equated points)	4.0300 points ✎	▃▂▂▃▃▃▄▅▅▇█▃ beats 6/247	avg 41.206 points	▇▇ 4.0%	blunt
Decision Index 0.4.1: private new domains (equated points)	7.0000 points ✎	▃█▃▃▂▆▃▃▄▄▄▄ beats 6/247	avg 38.861 points	▇ 2.2%	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/akash-kamat ↗

$ 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		= 18.4

not counted (2)

Decision Index 0.4.1	Decision Index 0.4.1: RouterBench (selected quality (quality objective)) [correct]	35.8%	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]	45.2%	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