whatmodel.sh / decision / vLLM-SR d3-mini (4B)

vLLM-SR d3-mini (4B)

+ contribute

vLLM-SR d3-mini (4B) is an open-weights decision model from vLLM Semantic Router. Its published size is 4.54B parameters. Licence: apache-2.0.

Catalogue identity notes

Exact Decision Index tested engine vllm-sr-d3-mini; reported model vLLM-SR d3-mini (4B). Type: full fine-tune; technique: autoregressive. No 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. Reviewed Ollaya local route added 2026-10-31: d3:mini, manifest 5c148d7803097eb575b7b90308496c3f8ec61890a06d5515098c4279aa7b7e39, matching the independent accuracy-run manifest. Native family implementation; published runtime precision label F32. Memory is measured peak GPU increase (11097 MiB, rounded upward to 10.9 GiB) on RTX 5090 during the short five-question triage workload, not download size or a maximum-context guarantee. Runtime installation is required; no fresh execution performed here. Historical benchmark adapter and input protocol still vary by result.

Model information sources

· Weights licence ↗ · Parameter count ↗

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

58.4±1.1#124 of 355 decision modelsfrom 68 benchmark results across 3 sources

$ how it compares

The decision models ranked either side of vLLM-SR d3-mini (4B), out of 357 with a score, then more from vLLM Semantic Router.

#	model	score	where	RAM	released
124	jev-d3-46k tinnel123	59±4	model files
125	Jebadiah 27B v2.1 (Ollaya Q4_K_M) frontier-infra	59±4	local	~16GB	2026-10
126	vLLM-SR d3-mini (4B) vLLM Semantic Router	58±1	local	~11GB
127	Seb-9B v2 ironbcc	58±1	model files
128	JPT-35B-A3B kirp	58±1	model files
	more from vLLM Semantic Router
8	vLLM-SR d3-ultra (Decision 3.0, 27B) vLLM Semantic Router	72±1	model files
9	vLLM-SR d3-turbo v1.1 (27B) vLLM Semantic Router	72±1	model files
25	vLLM-SR d3 (27B) vLLM Semantic Router	69±1	model files

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

$ what it's good at

tools & automation▁▃▁▁▁▂▃▄▆██▆ #110 of 250 · 5.3% of its score · 6 tests

Beats Senova-4B v2, SynACK Decide 26B-A4B v7, Tern-35B-A3B here, though they rank higher overall. Trails OneJev1.6-0.8B, TokenRhythm NeoHorse-Jev-4B, mmastrac 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: API-Bank (accuracy) submitted	83.9% ✎	▁▁▁▁▁▁▁▂▃▂█▂ beats 139/236	easy · avg 70%	▇ 1.6%	sharp · noisy
Decision Index 0.4.1: BFCL (case exact accuracy) submitted	95.5% ✎	▁▂▁▁▁▂▂▄▅▆█▁ beats 138/246	very easy · avg 92%	▇ 1.2%	sharp · maxed out · noisy
Decision Index 0.4.1: Home appliance simulator (case exact accuracy) submitted	37.5% ✎	█▁▁▃▄▄▆▄▁▂▁▃ beats 112/243	medium · avg 37%	▇ 0.9%	sharp · at its level · noisy
Decision Index 0.4.1: When2Call (accuracy) submitted	84.5% ✎	▁▁▁▂▃▅▅▆█▇█▃ beats 169/246	easy · avg 72%	▇ 0.8%	noisy
Decision Index 0.4.1: ToolRet (nDCG@10) submitted	64.4% ✎	▁▁▁▁▁·▁▁▁▄█▁ beats 123/246	medium · avg 60%	▇ 0.8%	blunt · at its level
Decision Index 0.4.1: RouterBench (selected quality (quality objective)) submitted	77.8% ✎	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▁▁▁·▁▁▁▁▃▃█▅ #135 of 258 · 5.2% of its score · 7 tests

Beats Quyet-1.0-Large, OneJev1.6-4B, Decider chat · Qwen3.6-27B (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	45.1% ✎	▁▁▁▁▁▁▁▂▂▄▇█ beats 152/246	medium · avg 38%	▇ 2.0%	blunt · at its level
Decision Index 0.4.1: Amazon ESCI (macro-F1) submitted	52.0% ✎	▁▁▁▁▁▁▂▂▃▃█▅ beats 119/246	medium · avg 47%	▇ 1.8%	at its level
Decision Index 0.4.1: HoVer (accuracy) submitted	81.5% ✎	▆▆▇▅▄▄▃▇▄█▆▄ beats 159/246	easy · avg 74%	▇ 0.5%	blunt
Decision Index 0.4.1: PhishNChips phishing decisions (accuracy) submitted	72.4% ✎	█▅▅▄▃▂▁▁▁▁·▁ beats 139/245	easy · avg 70%	▇ 0.4%	noisy
Decision Index 0.4.1: CLINC150+OOS (macro-F1) submitted	96.3% ✎	▁▁▁▁▁▁▁▂▃▆█▄ beats 224/238	easy · avg 83%	▇ 0.3%	sharp · noisy
Decision Index 0.4.1: BANKING77 (macro-F1) submitted	92.8% ✎	▁·▁▁▁▁▁▁▂█▇█ beats 226/238	easy · avg 79%	▇ 0.2%	noisy
Decision Index 0.4.1: SGD/SGD-X (macro-F1) submitted	47.3% ✎	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▂▂▂▃▂▅▆▃█▆▄▃ #47 of 256 · 17% of its score · 9 tests

Beats Ainvest-Decision-27B v1.0, Turion Jev 31B, MISA AI 2.0 Decision Plus · oak 1.1 (31B) here, though they rank higher overall. Trails Decider 35B-A3B, Sieve 9B v2, Clef-flash 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	67.9% ✎	▁▂▃▅▅█▂····▁ beats 143/246	medium · avg 60%	▇▇ 4.1%	at its level
Decision Index 0.4.1: NLI4CT (macro-F1) submitted	82.6% ✎	▁▁▁▂▂▃▃▄▇▇█▄ beats 183/246	easy · avg 74%	▇▇ 4.0%	typical
Decision Index 0.4.1: HellaSwag (accuracy) submitted	96.0% ✎	▂▄▂▂▄▇▃▇█▅▃▁ beats 209/246	very easy · avg 85%	▇ 3.4%	sharp · maxed out
Decision Index 0.4.1: iSarcasmEval (Sarcasm F1 · track A, English) submitted	50.2% ✎	▁·▁·▁▁▂▄▅█▇▄ beats 133/246	medium · avg 44%	▇ 1.9%	at its level
Decision Index 0.4.1: ACOS (per-review F1) submitted	17.4% ✎	▁··▂▃▃▅▇█▇▃▃ beats 91/246	hard · avg 22%	▇ 1.0%	at its level · noisy
Decision Index 0.4.1: ContractNLI (macro-F1) submitted	74.7% ✎	▁▁▁▁▁▁▁▂▂▇▇█ beats 113/246	easy · avg 71%	▇ 0.8%	noisy
Decision Index 0.4.1: VAST (macro-F1) submitted	70.5% ✎	▁▁▁▂▅▄▄▂▄▆█▄ beats 142/246	medium · avg 62%	▇ 0.8%	at its level
Decision Index 0.4.1: RAGTruth (F1 on hallucinated class) submitted	71.5% ✎	▁▁▁▁▁▁▁▁▁▃▆█ beats 104/246	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: FinEntity (macro-F1) submitted	93.5% ✎	▁···▁▁▁▁▃▆█▃ beats 208/246	easy · avg 84%	▇ 0.5%	noisy

arts & human taste▁▇▄▃▅▅█▅▄▂▁▁ #115 of 247 · 7.3% of its score · 6 tests

Beats Bobcat 1.3 (31B), Nace.AI Drex v2.0 (35B-A3B), decisio 0.6.0 (Gemma 4 31B) (decision readout) 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	76.1% ✎	▁▁▂▃▃▃▇▅█▅▃▁ beats 201/246	medium · avg 61%	▇ 3.0%	at its level
Decision Index 0.4.1: Humicroedit (accuracy) submitted	59.7% ✎	▁·▁▃▂▂▅█▄▂▁▁ beats 89/246	medium · avg 60%	▇ 1.5%	blunt · at its level
Decision Index 0.4.1: cfcolor (accuracy) submitted	61.8% ✎	▁········▁▇█ beats 124/246	medium · avg 60%	▇ 0.9%	blunt · at its level
Decision Index 0.4.1: BPoMP (accuracy) submitted	89.5% ✎	▁█▄▄▅█▆▅▄▁▁▁ beats 132/246	easy · avg 83%	▇ 0.8%	sharp · noisy
Decision Index 0.4.1: POP909-CL (accuracy) submitted	21.7% ✎	▄▅▅▃▆▅▃▅▂▅▃█ beats 132/237	hard · avg 21%	▇ 0.8%	sharp · at its level · noisy
Decision Index 0.4.1: Habermas Machine (accuracy) submitted	41.2% ✎	▁▂▄█▇▅▁▁▄▄▁▁ beats 86/246	medium · avg 47%	▇ 0.3%	blunt · at its level

knowledge & reasoning▁▂▂▂▂▃▃▅▅▆█▂ #104 of 263 · 38% of its score · 19 tests

Beats Senova-4B v2, Cerno 9B v1.1, OneJev1.6-4B here, though they rank higher overall. Trails Decider 35B-A3B, decisio 0.4.0 (Qwen3.6-35B-A3B) (decision readout), Solar Decide 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	58.9% ✎	▄▄▃▃▂▄▅▇██▅▁ beats 149/246	medium · avg 46%	▇▇▇ 8.2%	sharp · at its level
Decision Index 0.4.1: MMLU (accuracy) submitted	80.1% ✎	▁▄▄▃▃▂▅▇▅█▇▁ beats 159/246	easy · avg 70%	▇▇ 5.1%	sharp
Decision Index 0.4.1: BBH (accuracy) submitted	71.5% ✎	▄▆▄▄█▇▇▄▃▁▁▁ beats 142/246	easy · avg 66%	▇▇ 4.7%	typical
Decision Index 0.4.1: GSM8K (accuracy (4- and 10-choice)) submitted	69.6% ✎	▁··▄█▆▆▆▅▅▃▁ beats 185/246	medium · avg 45%	▇ 2.7%	sharp · at its level
Decision Index 0.4.1: GPQA Diamond (accuracy) submitted	44.9% ✎	▂▆▅▅▆█▆▃▁··▁ beats 151/246	medium · avg 39%	▇ 2.6%	blunt · at its level
Decision Index 0.4.1: ARC-Challenge (accuracy) submitted	95.5% ✎	▂▂▂▂▂▂▁▃▄▆█▆ beats 149/246	very easy · avg 90%	▇ 2.4%	sharp · maxed out
Decision Index 0.4.1: WinoGrande (accuracy) submitted	90.2% ✎	█▅█▅▄█▆▄▃▃▁▁ beats 200/246	easy · avg 78%	▇ 2.3%	typical
Ollaya · PubMedQA	71.6% ✎	▃▂▃▃·▂▃▃▆█▆▂ beats 14/25	easy · avg 69%	▇ 1.7%	typical

11 more tests

test	result	vs the others in its run	difficulty	counts	why
Ollaya · BoolQ	89.0% ✎	▁··▃▃▁▄▁▃█▇▄ beats 23/25	easy · avg 84%	▇ 1.2%	typical
Decision Index 0.4.1: SATA-Bench (case exact accuracy) submitted	19.0% ✎	▁▁▁▁▁▁▁▃▅█▂▁ beats 69/246	hard · avg 22%	▇ 1.2%	at its level
Decision Index 0.4.1: ARC-Easy (accuracy) submitted	98.9% ✎	▁▁▂▁▂▂▂▂▃▆█▅ beats 173/246	very easy · avg 96%	▇ 1.1%	sharp · maxed out · noisy
Decision Index 0.4.1: MuSR (accuracy) submitted	60.8% ✎	▁▁▂█▅▄▁▁▁▁▁▁ beats 140/246	medium · avg 60%	▇ 1.0%	blunt · at its level
Decision Index 0.4.1: CRUXEval (accuracy) submitted	64.6% ✎	██▄▃▃▄▁▁···▁ beats 152/246	medium · avg 58%	▇ 0.8%	at its level · noisy
Decision Index 0.4.1: CLadder (accuracy) submitted	72.8% ✎	▁███▆▅▃▃▁▁▁▁ beats 145/246	easy · avg 71%	▇ 0.7%	typical
Ollaya · SQuAD 2 answerability	74.3% ✎	▁▁▃▁▇█▃▃▄▃·▁ beats 8/25	easy · avg 80%	▇ 0.7%	blunt
Ollaya · VitaminC verification	82.6% ✎	▁···▃·▄█▃··▁ beats 23/25	easy · avg 78%	▇ 0.6%	blunt
Decision Index 0.4.1: ChessBench (accuracy) submitted	14.1% ✎	▄▇▄▄▄▆▄█▄▁▁▁ beats 106/246	hard · avg 16%	▇ 0.6%	blunt
Ollaya · MultiNLI	91.6% ✎	▂··▄██▂▆▆█▆▂ beats 22/25	very easy · avg 87%	▇ 0.4%	blunt
Decision Index 0.4.1: HLE (accuracy) submitted	10.8% ✎	▁▁▂▃▆█▇▅▅▃▁▁ beats 59/246	very hard · avg 13%	▇ 0.0%	blunt

vision▁·▃▂▃▃▇▄▄█▄▃ #40 of 79 · 10% of its score · 12 tests

Beats Surogate Rune 26B-A4B v3, Clef-omni here, though they rank higher overall. Trails Clef-flash here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index Vision 0.4 · Mind2Web (chance-corrected points) submitted	77.5300 points ✎	▁·▁▁▃▁▄▄▃▅█▂ beats 35/61	avg 63.423 points	▇ 1.3%	blunt
Decision Index Vision 0.4 · MMMU-Pro vision (chance-corrected points) submitted	32.7100 points ✎	▁▁▅▁▄▁▄█▄▃▇▇ beats 35/61	avg 22.797 points	▇ 1.2%	blunt
Decision Index Vision 0.4 · Winoground (chance-corrected points) submitted	81.5000 points ✎	▁·▁▁▁▁·▂▁▃▅█ beats 36/61	avg 66.987 points	▇ 1.1%	blunt
Decision Index Vision 0.4: private same skills (points)	58.6500 points ✎	▁▁▃▃▁▂▄▆▃▃▅█ beats 35/61	avg 48.067 points	▇ 1.1%	blunt
Decision Index Vision 0.4 · Moderation (Hateful Memes) (chance-corrected points) submitted	37.3500 points ✎	▆▃▃▆▄▄▇██▁▆▆ beats 32/61	avg 28.443 points	▇ 0.9%	blunt
Decision Index Vision 0.4 · RealWorldQA (chance-corrected points) submitted	57.8300 points ✎	▁·▂▂▁▃▃▅▅▃▂█ beats 23/61	avg 53.651 points	▇ 0.8%	blunt
Decision Index Vision 0.4 · CharXiv (chance-corrected points) submitted	73.9300 points ✎	▂▁▁▁▁▃▁▅▅▆██ beats 35/61	avg 62.137 points	▇ 0.8%	blunt
Decision Index Vision 0.4 · InfographicVQA (chance-corrected points) submitted	94.5400 points ✎	▁·▁▂▁▁▃▁▂▂▅█ beats 37/61	avg 83.859 points	▇ 0.6%	blunt

4 more tests

test	result	vs the others in its run	difficulty	counts	why
Decision Index Vision 0.4 · R-Bench-M (chance-corrected points) submitted	18.7800 points ✎	▂▄▇▆█▄▄▇▂··▁ beats 20/61	avg 19.676 points	▇ 0.6%	blunt
Decision Index Vision 0.4 · BLINK (chance-corrected points) submitted	53.4400 points ✎	▁···▁▂▂▂█▄▂▃ beats 29/61	avg 44.994 points	▇ 0.6%	blunt
Decision Index Vision 0.4 · CV-Bench (chance-corrected points) submitted	73.9400 points ✎	▁▁▁▁▂▁▁▁▄█▃▃ beats 27/61	avg 67.662 points	▇ 0.5%	blunt
Decision Index Vision 0.4 · KIE (CORD+FUNSD) (chance-corrected points) submitted	92.1900 points ✎	▁··▁▁▁▁▁▂▄█▁ beats 40/61	avg 85.184 points	▇ 0.4%	blunt

workflows▁▁▁▂▁▂▂▆█▄▁▁ #43 of 114 · 10% of its score · 9 tests

Beats Clef-omni, Jebadiah 27B v2.1 (Ollaya Q4_K_M) here, though they rank higher overall. Trails StartLux-Decision v1.1 2B, Solar Decide, openjev (DecisionBench ec4ff3df) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Ollaya · Typed decisions test	73.7% ✎	▄·▂▆▄▄▂·▄█▆▆ beats 22/22	medium · avg 64%	▇ 2.7%	at its level
Ollaya · SummEval relevance	51.7% ✎	▁··▁▂▁▅··▃██ beats 23/25	medium · avg 41%	▇ 1.7%	at its level
Ollaya · PAWS paraphrase	89.2% ✎	▂▂▂▂·▃█▆·▃·▁ beats 24/25	easy · avg 81%	▇ 1.5%	typical
Ollaya · Aegis 2 safety	80.4% ✎	▂·▁▁▂·▃▃▂█▃▁ beats 16/25	easy · avg 77%	▇ 1.3%	typical
Ollaya · MASSIVE German intent	86.9% ✎	▁····▁·▂▃▆█▃ beats 20/23	easy · avg 84%	▇ 1.1%	sharp
Ollaya · SummEval consistency	83.3% ✎	▁·▁▁··▁▁▁▁▄█ beats 18/25	easy · avg 75%	▇ 0.6%	sharp · noisy
Ollaya · HelpSteer 2 scoring	41.4% ✎	▁▁··▁·▁▃▇█▅▄ beats 19/23	medium · avg 38%	▇ 0.6%	blunt · at its level
Ollaya · MASSIVE English intent	88.6% ✎	▁·······▂▃█▃ beats 20/23	very easy · avg 86%	▇ 0.6%	noisy

1 more test

test	result	vs the others in its run	difficulty	counts	why
Ollaya · Civil Comments	93.3% ✎	▂▂▂·▂▅▅▆▃▅·█ beats 22/23	very easy · avg 86%	▇ 0.1%	blunt · noisy

other tests2 not in a task · 6.7% of its score

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: private same skills (equated points)	51.4500 points ✎	▂▂▂▃▃▃▄▅▅▇█▃ beats 151/247	avg 41.206 points	▇▇ 4.3%	blunt
Decision Index 0.4.1: private new domains (equated points)	47.7500 points ✎	▃█▃▃▂▆▃▃▄▄▄▄ beats 151/247	avg 38.861 points	▇ 2.4%	blunt

$ where to run it

Other ways to run it

Run it as published · needs 10.9 GB, NVIDIA only · from github.com/ollaya-dev
ollaya run d3:mini --preset triage "The customer requests a refund."

Download model files ↗
Published model files verified. This is a verified repository snapshot; the benchmark does not establish an identical weight revision. A matching Ollaya runtime manifest and published short-workload GPU memory measurement were reviewed separately; see local variants and identity_note. The command was not executed in this review. Source · checked 2026-10-31.

Model repository and documentation: huggingface.co/vllm-sr ↗

$ how the score was worked out · Decision Index 0.4.1 75%, Ollaya · original runs 15%, Decision Index Vision 0.4 10%

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	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 75%	CC BY 4.0
Ollaya · original runs	14	▇▇▇▇ 15%	Apache-2.0
Decision Index Vision 0.4 · submitted; independent execution unverified	12	▇▇▇ 10%	CC BY 4.0
prior · pull toward the middle		▇ 1%
weighted average		= 58.4

not counted (2)

Decision Index 0.4.1	Decision Index 0.4.1: RouterBench (selected quality (quality objective)) [correct]	77.8%	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]	47.3%	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