whatmodel.sh / decision / Nari Decision 4B v3.2

Nari Decision 4B v3.2

+ contribute

Nari Decision 4B v3.2 is an open-weights decision model from blackmallangcow, released 18 October 2026. Its published size is 4.22B parameters. Licence: apache-2.0.

Catalogue identity notes

Exact Decision Index tested engine nari-decision-4b-v3.2; reported model Nari Decision 4B v3.2. Type: full fine-tune; 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

Release date ↗ · Weights licence ↗ · Parameter count ↗

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

49.2±1.2#212 of 355 decision modelsfrom 42 benchmark results across 1 source

$ how it compares

The decision models ranked either side of Nari Decision 4B v3.2, out of 357 with a score.

#	model	score	where	RAM	released
212	FlyMy.AI Decision 4B v1.6 FlyMy.AI	50±1	model files
213	TinyJev-4B (JevBench mini) AnkitAI	50±2	source code
214	Nari Decision 4B v3.2 blackmallangcow	49±1	model files		2026-10-18
215	Hopper (G) 1.2 Qwen3.5-4B LoRA HopitAI	49±1	model files
216	vLLM-SR d3-nano (2B) vLLM Semantic Router	49±1	local	~8.2GB

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

$ what it's good at

tools & automation▁▃▁▁▁▂▃▄▆██▆ #151 of 250 · 8.1% of its score · 6 tests

Beats Cerno 4B v1.1, Trio-Spark DI (9B), Decider 35B-A3B here, though they rank higher overall. Trails JevK5 Lite, Gevva E2B, 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: BFCL (case exact accuracy) submitted	92.4% ✎	▁▂▁▁▁▂▂▄▅▆█▁ beats 98/246	very easy · avg 92%	▇ 2.7%	sharp · noisy
Decision Index 0.4.1: API-Bank (accuracy) submitted	50.4% ✎	▁▁▁▁▁▁▁▂▃▂█▂ beats 51/236	easy · avg 70%	▇ 2.5%	sharp · at its level · noisy
Decision Index 0.4.1: When2Call (accuracy) submitted	72.3% ✎	▁▁▁▂▃▅▅▆█▇█▃ beats 103/246	easy · avg 72%	▇ 1.1%	at its level · noisy
Decision Index 0.4.1: Home appliance simulator (case exact accuracy) submitted	42.0% ✎	█▁▁▃▄▄▆▄▁▂▁▃ beats 116/243	medium · avg 37%	▇ 0.9%	sharp · at its level · noisy
Decision Index 0.4.1: ToolRet (nDCG@10) submitted	68.2% ✎	▁▁▁▁▁·▁▁▁▄█▁ beats 202/246	medium · avg 60%	▇ 0.9%	blunt · at its level
Decision Index 0.4.1: RouterBench (selected quality (quality objective)) submitted	79.0% ✎	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▁▁▁·▁▁▁▁▃▃█▅ #138 of 258 · 5.9% of its score · 7 tests

Beats Quyet-1.0-Large, Decider chat · Qwen3.6-27B (decision readout), Xor 1.2 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	31.9% ✎	▁▁▁▁▁▁▁▂▃▄▇█ beats 58/246	medium · avg 38%	▇ 2.1%	blunt · at its level
Decision Index 0.4.1: Amazon ESCI (macro-F1) submitted	52.2% ✎	▁▁▁▁▁▁▂▂▃▃█▅ beats 121/246	medium · avg 47%	▇ 2.0%	at its level
Decision Index 0.4.1: HoVer (accuracy) submitted	87.0% ✎	▆▆▇▅▄▄▃▇▄█▆▄ beats 208/246	easy · avg 74%	▇ 0.6%	blunt · at its level
Decision Index 0.4.1: PhishNChips phishing decisions (accuracy) submitted	73.9% ✎	█▅▅▄▃▂▁▁▁▁·▁ beats 144/245	easy · avg 70%	▇ 0.5%	at its level · noisy
Decision Index 0.4.1: CLINC150+OOS (macro-F1) submitted	89.7% ✎	▁▁▁▁▁▁▁▂▃▆█▄ beats 122/238	easy · avg 83%	▇ 0.5%	sharp · noisy
Decision Index 0.4.1: BANKING77 (macro-F1) submitted	81.9% ✎	▁·▁▁▁▁▁▁▂█▇█ beats 112/238	easy · avg 79%	▇ 0.3%	noisy
Decision Index 0.4.1: SGD/SGD-X (macro-F1) submitted	17.5% ✎	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▂▂▂▃▂▅▆▃█▆▄▃ #193 of 256 · 24% of its score · 9 tests

Trails DecisionTune 1.0, vera-spark, JEV FLAN-T5 Large public v2 (342M) 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	74.3% ✎	▂▄▂▂▄▇▃▇█▅▃▁ beats 61/246	very easy · avg 85%	▇▇▇ 6.7%	sharp
Decision Index 0.4.1: NLI4CT (macro-F1) submitted	76.3% ✎	▁▁▁▂▂▃▃▄▇▇█▄ beats 110/246	easy · avg 74%	▇▇ 5.4%	at its level
Decision Index 0.4.1: ANLI (macro-F1) submitted	58.2% ✎	▁▂▃▅▅█▂····▁ beats 99/246	medium · avg 60%	▇▇ 5.0%	at its level
Decision Index 0.4.1: iSarcasmEval (Sarcasm F1 · track A, English) submitted	45.7% ✎	▁·▁·▁▁▂▄▅█▇▄ beats 111/246	medium · avg 44%	▇ 2.0%	at its level
Decision Index 0.4.1: ContractNLI (macro-F1) submitted	66.1% ✎	▁▁▁▁▁▁▁▂▃▇▇█ beats 63/246	easy · avg 71%	▇ 1.1%	at its level · noisy
Decision Index 0.4.1: VAST (macro-F1) submitted	49.1% ✎	▁▁▁▂▅▄▄▂▄▆█▄ beats 74/246	medium · avg 62%	▇ 1.0%	at its level
Decision Index 0.4.1: ACOS (per-review F1) submitted	22.3% ✎	▁··▂▃▃▅▇█▇▃▃ beats 123/246	hard · avg 22%	▇ 0.9%	noisy
Decision Index 0.4.1: RAGTruth (F1 on hallucinated class) submitted	71.7% ✎	▁▁▁▁▁▁▁▁▁▃▆█ beats 107/246	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	82.6% ✎	▁···▁▁▁▁▃▆█▃ beats 83/246	easy · avg 84%	▇ 0.7%	noisy

arts & human taste▁▇▄▃▅▅█▅▄▂▁▁ #166 of 247 · 8.8% of its score · 6 tests

Beats PurpleMIST-Flash-1.0, Jev-Omni, vLLM-SR Decision 2.0 Nox 4B here, though they rank higher overall. Trails Crivo-17M, Crivo-32M v2, Crivo-68M 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	54.4% ✎	▁▁▂▃▃▃▇▅█▅▃▁ beats 71/246	medium · avg 61%	▇▇ 3.8%	at its level
Decision Index 0.4.1: Humicroedit (accuracy) submitted	59.6% ✎	▁·▁▃▂▂▅█▄▂▁▁ beats 87/246	medium · avg 60%	▇ 1.7%	blunt · at its level
Decision Index 0.4.1: BPoMP (accuracy) submitted	77.2% ✎	▁█▄▄▅█▆▅▄▁▁▁ beats 81/246	easy · avg 83%	▇ 1.4%	sharp · noisy
Decision Index 0.4.1: cfcolor (accuracy) submitted	63.1% ✎	▁········▁▇█ beats 146/246	medium · avg 60%	▇ 1.0%	blunt · at its level
Decision Index 0.4.1: POP909-CL (accuracy) submitted	1.2% ✎	▄▅▅▃▆▅▃▅▂▅▃█ beats 20/237	hard · avg 21%	▇ 0.6%	sharp · noisy
Decision Index 0.4.1: Habermas Machine (accuracy) submitted	39.6% ✎	▁▂▄█▇▅▁▁▄▄▁▁ beats 59/246	medium · avg 47%	▇ 0.3%	blunt · at its level

knowledge & reasoning▁▂▂▂▂▃▃▅▅▆█▂ #154 of 263 · 46% of its score · 14 tests

Beats StartLux-Decision v1.1 2B, Solomon v1.1, Standard One 3B SH (v4) here, though they rank higher overall. Trails Ryotide Qwen v0.3.1 (decision readout), Interfaze lev Qwen3.5-4B LoRA, razorback16 diffgemma (decision readout) 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	49.2% ✎	▄▄▃▃▂▄▅▇██▅▁ beats 109/246	medium · avg 46%	▇▇▇▇ 9.2%	sharp · at its level
Decision Index 0.4.1: MMLU (accuracy) submitted	72.4% ✎	▁▄▄▃▃▂▅▇▅█▇▁ beats 106/246	easy · avg 70%	▇▇▇ 7.4%	sharp · at its level
Decision Index 0.4.1: BBH (accuracy) submitted	62.7% ✎	▄▆▄▄█▇▇▄▃▁▁▁ beats 94/246	easy · avg 66%	▇▇▇ 6.4%	at its level
Decision Index 0.4.1: ARC-Challenge (accuracy) submitted	91.8% ✎	▂▂▂▂▂▂▁▃▄▆█▆ beats 104/246	very easy · avg 90%	▇▇ 5.6%	sharp
Decision Index 0.4.1: WinoGrande (accuracy) submitted	80.7% ✎	█▅█▅▄█▆▄▃▃▁▁ beats 128/246	easy · avg 78%	▇ 3.3%	typical
Decision Index 0.4.1: ARC-Easy (accuracy) submitted	97.0% ✎	▁▁▂▁▂▂▂▂▃▆█▅ beats 103/246	very easy · avg 96%	▇ 3.3%	sharp · maxed out · noisy
Decision Index 0.4.1: GSM8K (accuracy (4- and 10-choice)) submitted	38.9% ✎	▁··▄█▆▆▆▅▅▃▁ beats 104/246	medium · avg 45%	▇ 3.1%	sharp · at its level
Decision Index 0.4.1: GPQA Diamond (accuracy) submitted	45.9% ✎	▂▆▅▅▆█▆▃▁··▁ beats 159/246	medium · avg 39%	▇ 2.7%	blunt · 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	58.6% ✎	▁▁▂█▅▄▁▁▁▁▁▁ beats 117/246	medium · avg 60%	▇ 1.1%	blunt · at its level
Decision Index 0.4.1: SATA-Bench (case exact accuracy) submitted	24.6% ✎	▁▁▁▁▁▁▁▂▅█▂▁ beats 103/246	hard · avg 22%	▇ 1.1%	typical
Decision Index 0.4.1: CRUXEval (accuracy) submitted	49.8% ✎	██▄▃▃▄▁▁···▁ beats 109/246	medium · avg 58%	▇ 1.0%	at its level · noisy
Decision Index 0.4.1: CLadder (accuracy) submitted	61.8% ✎	▁███▆▅▃▃▁▁▁▁ beats 70/246	easy · avg 71%	▇ 0.9%	at its level
Decision Index 0.4.1: ChessBench (accuracy) submitted	10.7% ✎	▄▇▄▄▄▆▄█▄▁▁▁ beats 69/246	hard · avg 16%	▇ 0.5%	blunt
Decision Index 0.4.1: HLE (accuracy) submitted	10.4% ✎	▁▁▂▃▆█▇▅▅▃▁▁ beats 38/246	very hard · avg 13%	▇ 0.0%	blunt

other tests2 not in a task · 7.3% of its score

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: private same skills (equated points)	42.8300 points ✎	▂▂▂▃▃▃▄▅▅▇█▃ beats 114/247	avg 41.206 points	▇▇ 4.7%	blunt
Decision Index 0.4.1: private new domains (equated points)	37.9400 points ✎	▃█▃▃▂▆▃▃▄▄▄▄ beats 115/247	avg 38.861 points	▇ 2.6%	blunt

$ where to run it

Download model files ↗
Published model files verified. 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/blackmallangcow ↗

$ 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		= 49.2

not counted (2)

Decision Index 0.4.1	Decision Index 0.4.1: RouterBench (selected quality (quality objective)) [correct]	79.0%	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]	17.5%	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