whatmodel.sh / decision / Span-02 (Gemma 4 26B-A4B)

Span-02 (Gemma 4 26B-A4B)

+ contribute

Span-02 (Gemma 4 26B-A4B) is an open-weights decision model from Respan. Its published size is 25.81B parameters. Licence: apache-2.0.

Catalogue identity notes

Exact Decision Index tested engine span-02; reported model span-02 (Gemma 4 26B-A4B). 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. Canonical HF API resolves lihe07/span-02 to respanai/span-02 at the same 63cbdd774391f1b7d9bc3829179efe996d0191ce revision. Maker attribution corrected to Respan, confirmed by the model card citation and official span-02 docs. Native serving uses adaptive reasoning; historical per-evaluator budgets remain in result notes.

Model information sources

· Weights licence ↗ · Parameter count ↗

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

65.9±1.1#50 of 355 decision modelsfrom 76 benchmark results across 4 sources

$ how it compares

The decision models ranked either side of Span-02 (Gemma 4 26B-A4B), out of 357 with a score, then more from Respan.

#	model	score	where	released
48	allec-oavg (Gemma 4 31B) plantwaterer	66±1	model files
49	Jev 1.13 TypeSafe	66±1	api $0.042/M input tokens	2026-09-15
50	Span-02 (Gemma 4 26B-A4B) Respan	66±1	api price unknown
51	RSI-Jev v6.1 VL 27B (effort auto) shgao	66±1	model files	2026-10-08
52	KnowLine 9B Gen1 PelaAI	66±1	model files
	more from Respan
207	Span-01 Pro Respan	50±9	api price unknown	2026-09-24

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

$ what it's good at

tools & automation▁▃▁▁▁▂▃▄▆██▆ #86 of 250 · 3.3% of its score · 6 tests

Beats Sieve 27B v2, Jev 1.13 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	80.9% ✎	▁▁▁▁▁▁▁▂▃▂█▂ beats 122/236	easy · avg 70%	▇ 1.0%	sharp · noisy
Decision Index 0.4.1: ToolRet (nDCG@10) submitted	66.1% ✎	▁▁▁▁▁·▁▁▁▄█▁ beats 164/246	medium · avg 60%	▇ 0.6%	blunt · at its level
Decision Index 0.4.1: Home appliance simulator (case exact accuracy) submitted	100.0% ✎	█▁▁▃▄▄▆▄▁▂▁▃ beats 239/243	medium · avg 37%	▇ 0.6%	sharp · noisy
Decision Index 0.4.1: BFCL (case exact accuracy) submitted	96.2% ✎	▁▂▁▁▁▂▂▄▅▆█▁ beats 156/246	very easy · avg 92%	▇ 0.5%	sharp · maxed out · noisy
Decision Index 0.4.1: When2Call (accuracy) submitted	86.3% ✎	▁▁▁▂▃▅▅▆█▇█▃ beats 180/246	easy · avg 72%	▇ 0.5%	noisy
Decision Index 0.4.1: RouterBench (selected quality (quality objective)) submitted	79.9% ✎	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▁▁▁·▁▁▁▁▃▃█▅ #58 of 258 · 11% of its score · 13 tests

Beats AutoTrust-JEV-27B v1.1, d1.1 (hosted), oenogarum (Gemma 4 31B) here, though they rank higher overall. Trails StartLux-Decision v1.1 0.8B, Standard One 8B SH (v4), StartLux-Decision v1.1 2B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
System One Bench: MIRACL pairwise (Arabic)	91.5% ✎	▂···▁··▃▄▂█▇ beats 28/29	easy · avg 84%	▇ 2.2%	typical
System One Bench: MASSIVE intent (Arabic)	86.0% ✎	▄·▁·▁▃▃▃█▄▇▇ beats 17/29	easy · avg 76%	▇ 1.9%	typical
Decision Index 0.4.1: BRIGHT (nDCG@10) submitted	46.2% ✎	▁▁▁▁▁▁▁▂▂▄▇█ beats 169/246	medium · avg 38%	▇ 1.7%	blunt · at its level
Decision Index 0.4.1: Amazon ESCI (macro-F1) submitted	58.7% ✎	▁▁▁▁▁▁▂▂▃▃█▅ beats 200/246	medium · avg 47%	▇ 1.5%	at its level
System One Bench: MIRACL rerank (Arabic)	88.5% ✎	▁▂·▁▁·▂·▄▄█▇ beats 21/29	easy · avg 79%	▇ 1.3%	typical
System One Bench: MASSIVE scenario (Arabic)	71.0% ✎	▃▁·▄▃▂█▂▁▁·▃ beats 16/29	medium · avg 60%	▇ 0.9%	at its level
Decision Index 0.4.1: HoVer (accuracy) submitted	80.7% ✎	▆▆▇▅▄▄▃▇▄█▆▄ beats 150/246	easy · avg 74%	▇ 0.3%	blunt
Decision Index 0.4.1: PhishNChips phishing decisions (accuracy) submitted	81.2% ✎	█▅▅▄▃▂▁▁▁▁·▁ beats 176/245	easy · avg 70%	▇ 0.3%	noisy

5 more tests

test	result	vs the others in its run	difficulty	counts	why
System One Bench: AG News (English)	86.5% ✎	▁···▅▃▅▂▆█▂▅ beats 5/29	easy · avg 88%	▇ 0.2%	blunt
Decision Index 0.4.1: BANKING77 (macro-F1) submitted	85.6% ✎	▁·▁▁▁▁▁▁▂█▇█ beats 138/238	easy · avg 79%	▇ 0.2%	noisy
Decision Index 0.4.1: CLINC150+OOS (macro-F1) submitted	92.6% ✎	▁▁▁▁▁▁▁▂▃▆█▄ beats 169/238	easy · avg 83%	▇ 0.2%	sharp · maxed out · noisy
System One Bench: Banking77 (English)	82.0% ✎	█▅▆█▅▃▃▃···▂ beats 5/26	very easy · avg 87%	▇ 0.0%	blunt · noisy
Decision Index 0.4.1: SGD/SGD-X (macro-F1) submitted	66.7% ✎	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▂▂▂▃▂▅▆▃█▆▄▃ #68 of 256 · 18% of its score · 18 tests

Beats Turion Jev 31B, AJev lm6q (Gemma 4 31B), oenogarum (Gemma 4 31B) here, though they rank higher overall. Trails Sieve 4B, Decider 35B-A3B, Sieve 9B v2 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	78.1% ✎	▁▂▃▅▅█▃····▁ beats 224/246	medium · avg 60%	▇ 3.1%	typical
Decision Index 0.4.1: NLI4CT (macro-F1) submitted	82.3% ✎	▁▁▁▂▂▃▃▄▇▇█▄ beats 175/246	easy · avg 74%	▇ 2.9%	typical
Decision Index 0.4.1: HellaSwag (accuracy) submitted	94.3% ✎	▂▄▂▂▄▇▃▇█▅▃▁ beats 180/246	very easy · avg 85%	▇ 1.7%	sharp · maxed out
System One Bench: BoolQ (English)	90.0% ✎	▁▁▁▃·▁▃▃▇▇█▅ beats 15/29	easy · avg 86%	▇ 1.5%	typical
Decision Index 0.4.1: iSarcasmEval (Sarcasm F1 · track A, English) submitted	53.9% ✎	▁·▁·▁▁▂▄▅█▇▄ beats 151/246	medium · avg 44%	▇ 1.5%	at its level
System One Bench: OSACT4 offensive (Arabic)	97.0% ✎	▁······▁▃█▇▄ beats 29/29	easy · avg 84%	▇ 1.3%	sharp
Decision Index 0.4.1: ACOS (per-review F1) submitted	34.2% ✎	▁··▂▃▃▅▇█▇▃▃ beats 180/246	hard · avg 22%	▇ 1.0%	at its level · noisy
System One Bench: SST-5 (English)	57.5% ✎	▃·▁▁▃·▇▄▄▄█▅ beats 19/29	medium · avg 51%	▇ 0.9%	blunt · at its level

10 more tests

test	result	vs the others in its run	difficulty	counts	why
System One Bench: SST-2 (English)	95.5% ✎	▁·▃▁▂▂▃▅█·▂▁ beats 18/29	easy · avg 92%	▇ 0.9%	typical
Decision Index 0.4.1: VAST (macro-F1) submitted	77.3% ✎	▁▁▁▂▅▄▄▂▄▆█▄ beats 189/246	medium · avg 62%	▇ 0.6%	at its level
Decision Index 0.4.1: ContractNLI (macro-F1) submitted	78.1% ✎	▁▁▁▁▁▁▁▂▂▇▇█ beats 142/246	easy · avg 71%	▇ 0.6%	noisy
System One Bench: XNLI Arabic (Arabic)	66.0% ✎	▄▁▃▁▄▇█▅▃▁·▃ beats 7/29	medium · avg 68%	▇ 0.5%	at its level
Decision Index 0.4.1: RAGTruth (F1 on hallucinated class) submitted	80.8% ✎	▁▁▁▁▁▁▁▁▁▃▆█ beats 187/246	easy · avg 66%	▇ 0.4%	noisy
System One Bench: XNLI French (French)	73.3% ✎	▂▅·▃▁▅▅█▃▁·▃ beats 6/29	medium · avg 76%	▇ 0.4%	blunt · at its level
System One Bench: Tweet sentiment (Arabic)	76.0% ✎	▁······▁▁▄█▄ beats 25/29	medium · avg 62%	▇ 0.4%	blunt · at its level
Decision Index 0.4.1: FinEntity (macro-F1) submitted	92.5% ✎	▁···▁▁▁▁▃▆█▃ beats 181/246	easy · avg 84%	▇ 0.3%	maxed out · noisy
System One Bench: XNLI English (English)	83.3% ✎	▃·▁·▃▇█▄█▃▃▁ beats 6/29	easy · avg 85%	▇ 0.1%	blunt · noisy
System One Bench: DAIR Emotion (English)	56.0% ✎	██▃·····▁··▁ beats 7/29	medium · avg 59%	▇ 0.0%	blunt · at its level

arts & human taste▁▇▄▃▅▅█▅▄▂▁▁ #57 of 247 · 5.7% of its score · 6 tests

Beats OneJev1.6-27B, mogan-decision-31B-it, Cerno 27B here, though they rank higher overall. Trails Kas 4B, baselabs joey-1 (2B), Clef-flash 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	73.7% ✎	▁▁▂▃▃▃▇▅█▅▃▁ beats 176/246	medium · avg 61%	▇ 2.2%	typical
Decision Index 0.4.1: Humicroedit (accuracy) submitted	62.1% ✎	▁·▁▃▂▂▅█▄▂▁▁ beats 151/246	medium · avg 60%	▇ 1.3%	blunt · at its level
Decision Index 0.4.1: POP909-CL (accuracy) submitted	63.9% ✎	▄▅▅▃▆▅▃▅▂▅▃█ beats 187/237	hard · avg 21%	▇ 0.7%	sharp · at its level · noisy
Decision Index 0.4.1: cfcolor (accuracy) submitted	62.2% ✎	▁········▁▇█ beats 130/246	medium · avg 60%	▇ 0.7%	blunt · at its level
Decision Index 0.4.1: BPoMP (accuracy) submitted	95.0% ✎	▁█▄▄▅█▆▅▄▁▁▁ beats 190/246	easy · avg 83%	▇ 0.5%	sharp · maxed out · noisy
Decision Index 0.4.1: Habermas Machine (accuracy) submitted	47.5% ✎	▁▂▄█▇▅▁▁▄▄▁▁ beats 164/246	medium · avg 47%	▇ 0.2%	blunt · at its level

knowledge & reasoning▁▂▂▂▂▃▃▅▅▆█▂ #39 of 263 · 34% of its score · 15 tests

Beats vera-large, KnowLine 26B-A4B Gen2.5, OneJev1.6-27B here, though they rank higher overall. Trails Solar Decide, Clef-omni, GPT-6 Luna Decisions here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Etsabary · 1342 text decisions	84.0% ✎	▁··▂·▃▅▁▃▂▂█ beats 22/24	medium · avg 47%	▇▇▇▇ 10%	sharp
Decision Index 0.4.1: MMLU-Pro (accuracy) submitted	73.0% ✎	▄▄▃▃▂▄▅▇██▅▁ beats 229/246	medium · avg 46%	▇▇▇ 6.3%	sharp · at its level
Decision Index 0.4.1: BBH (accuracy) submitted	87.6% ✎	▄▆▄▄█▇▇▄▃▁▁▁ beats 224/246	easy · avg 66%	▇ 3.3%	typical
Decision Index 0.4.1: MMLU (accuracy) submitted	86.1% ✎	▁▄▄▃▃▂▅▇▅█▇▁ beats 199/246	easy · avg 70%	▇ 3.3%	sharp
Decision Index 0.4.1: GPQA Diamond (accuracy) submitted	46.9% ✎	▂▆▅▅▆█▆▃▁··▁ beats 166/246	medium · avg 39%	▇ 2.3%	blunt · at its level
Decision Index 0.4.1: GSM8K (accuracy (4- and 10-choice)) submitted	91.2% ✎	▁··▄█▆▆▆▅▅▃▃ beats 244/246	medium · avg 45%	▇ 2.1%	sharp · at its level
Decision Index 0.4.1: WinoGrande (accuracy) submitted	88.6% ✎	█▅█▅▄█▆▄▃▃▁▁ beats 190/246	easy · avg 78%	▇ 1.5%	typical
Decision Index 0.4.1: SATA-Bench (case exact accuracy) submitted	35.0% ✎	▁▁▁▁▁▁▁▂▅█▂▁ beats 206/246	hard · avg 22%	▇ 1.2%	at its level

7 more tests

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: ARC-Challenge (accuracy) submitted	97.4% ✎	▂▂▂▂▂▂▁▃▄▆█▆ beats 218/246	very easy · avg 90%	▇ 1.0%	sharp · maxed out
Decision Index 0.4.1: MuSR (accuracy) submitted	66.5% ✎	▁▁▂█▅▄▁▁▁▁▁▁ beats 194/246	medium · avg 60%	▇ 0.8%	blunt · at its level
Decision Index 0.4.1: CRUXEval (accuracy) submitted	82.3% ✎	██▄▃▃▄▁▁···▁ beats 205/246	medium · avg 58%	▇ 0.6%	at its level · noisy
Decision Index 0.4.1: CLadder (accuracy) submitted	80.5% ✎	▁███▆▅▃▃▁▁▁▁ beats 177/246	easy · avg 71%	▇ 0.5%	typical
Decision Index 0.4.1: ChessBench (accuracy) submitted	18.1% ✎	▄▇▄▄▄▆▄█▄▁▁▁ beats 154/246	hard · avg 16%	▇ 0.5%	blunt
Decision Index 0.4.1: ARC-Easy (accuracy) submitted	99.0% ✎	▁▁▂▁▂▂▂▂▃▆█▅ beats 182/246	very easy · avg 96%	▇ 0.4%	sharp · maxed out · noisy
Decision Index 0.4.1: HLE (accuracy) submitted	12.4% ✎	▁▁▂▃▆█▇▅▅▃▁▁ beats 123/246	very hard · avg 13%	▇ 0.0%	blunt

vision▁·▃▂▃▃▇▄▄█▄▃ #17 of 79 · 20% of its score · 17 tests

Beats StartLux-Decision v1.1 27B, Nace.AI Drex Max v1 (27B), AutoTrust-JEV-27B v1.1 here, though they rank higher overall. Trails Clef-flash, Mercury Decide (free endpoint), Clef here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
System One Bench: Image object count	88.5% ✎	▃····▃█▃▅·█▃ beats 4/11	easy · avg 71%	▇ 2.8%	sharp
System One Bench: SST-2 screenshot	92.5% ✎	█···███·█·▄▄ beats 2/11	easy · avg 87%	▇ 2.3%	sharp
System One Bench: Visual yes/no questions	87.0% ✎	▂······▃█▃·▃ beats 3/11	medium · avg 80%	▇ 2.2%	typical
System One Bench: Arabic sentiment screenshot	70.0% ✎	▄··▄··▂···▆█ beats 11/11	very hard · avg 37%	▇ 2.2%	sharp · at its level · noisy
System One Bench: Oxford pets classification	92.5% ✎	███·█▄▄·▄··▄ beats 11/11	easy · avg 77%	▇ 1.8%	typical
Decision Index Vision 0.4 · Mind2Web (chance-corrected points) submitted	73.0000 points ✎	▁·▁▁▃▁▄▄▃▅█▂ beats 27/61	avg 63.423 points	▇ 1.1%	blunt
Decision Index Vision 0.4 · MMMU-Pro vision (chance-corrected points) submitted	42.8500 points ✎	▁▁▅▁▄▁▄█▄▃▇▇ beats 47/61	avg 22.797 points	▇ 1.0%	blunt
Decision Index Vision 0.4 · Winoground (chance-corrected points) submitted	85.2500 points ✎	▁·▁▁▁▁·▂▁▃▅█ beats 50/61	avg 66.987 points	▇ 0.9%	blunt

9 more tests

test	result	vs the others in its run	difficulty	counts	why
Decision Index Vision 0.4: private same skills (points)	60.8500 points ✎	▁▁▃▃▁▂▄▆▃▃▅█ beats 40/61	avg 48.067 points	▇ 0.9%	blunt
Decision Index Vision 0.4 · Moderation (Hateful Memes) (chance-corrected points) submitted	37.3500 points ✎	▆▃▃▆▄▄▇██▁▆▆ beats 31/61	avg 28.443 points	▇ 0.8%	blunt
Decision Index Vision 0.4 · RealWorldQA (chance-corrected points) submitted	59.7700 points ✎	▁·▂▂▁▃▃▅▅▃▂█ beats 29/61	avg 53.651 points	▇ 0.7%	blunt
Decision Index Vision 0.4 · CharXiv (chance-corrected points) submitted	78.2000 points ✎	▂▁▁▁▁▃▁▅▅▆██ beats 43/61	avg 62.137 points	▇ 0.7%	blunt
Decision Index Vision 0.4 · InfographicVQA (chance-corrected points) submitted	95.6300 points ✎	▁·▁▂▁▁▃▁▂▂▅█ beats 41/61	avg 83.859 points	▇ 0.6%	blunt
Decision Index Vision 0.4 · R-Bench-M (chance-corrected points) submitted	35.5700 points ✎	▂▄▇▆█▄▄▇▂··▁ beats 51/61	avg 19.676 points	▇ 0.6%	blunt
Decision Index Vision 0.4 · BLINK (chance-corrected points) submitted	63.9100 points ✎	▁···▁▂▂▂█▄▂▃ beats 50/61	avg 44.994 points	▇ 0.5%	blunt
Decision Index Vision 0.4 · CV-Bench (chance-corrected points) submitted	75.6400 points ✎	▁▁▁▁▂▁▁▁▄█▃▃ beats 33/61	avg 67.662 points	▇ 0.4%	blunt
Decision Index Vision 0.4 · KIE (CORD+FUNSD) (chance-corrected points) submitted	91.9900 points ✎	▁··▁▁▁▁▁▂▄█▁ beats 37/61	avg 85.184 points	▇ 0.3%	blunt

workflows▁▁▁▂▁▂▂▆█▄▁▁ #30 of 114 * · 2.1% of its score · 1 test

Beats d1.1 (hosted) here, though they rank higher overall. Trails StartLux-Decision v1.1 2B, openjev-sglang (DecisionBench 1355db6a), decider-35b-a3b (DecisionBench d9783c58) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
System One Bench: Typed decisions (English)	72.1% ✎	▁▁·▁▁▁▅▄▇██▁ beats 18/29	medium · avg 63%	▇ 2.1%	blunt · at its level

other tests2 not in a task · 5.8% of its score

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: private same skills (equated points)	58.9500 points ✎	▂▂▂▃▃▃▄▅▅▇█▃ beats 204/247	avg 41.206 points	▇▇ 3.8%	blunt
Decision Index 0.4.1: private new domains (equated points)	59.2600 points ✎	▃█▃▃▂▆▃▃▄▄▄▄ beats 183/247	avg 38.861 points	▇ 2.1%	blunt

$ where to run it

Through an API

api · OpenRouter · price unknown · openrouter.ai ↗

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/respanai ↗

$ how the score was worked out · Decision Index 0.4.1 55%, System One Bench (Ali Alhashi… 26%, Etsabary · decision benchmarks 10% and 1 more

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	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 55%	CC BY 4.0
System One Bench (Ali Alhashimi)	21	▇▇▇▇▇▇▇▇ 26%	CC-BY-4.0
Etsabary · decision benchmarks	1	▇▇▇ 10%	MIT
Decision Index Vision 0.4 · submitted; independent execution unverified	12	▇▇▇ 9%	CC BY 4.0
prior · pull toward the middle		▇ 0%
weighted average		= 65.9

not counted (2)

Decision Index 0.4.1	Decision Index 0.4.1: RouterBench (selected quality (quality objective)) [correct]	79.9%	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]	66.7%	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