Clef-omni

+ contribute

Clef-omni is an open-weights decision model from Cloudflare, released October 2026. Its published size is 35.26B parameters. Licence: apache-2.0.

Catalogue identity notes

Exact Decision Index 0.4.1 text-board engine clef-omni. Produces probabilities over supplied choices, not conversational text. Only public per-benchmark results are used; no Full index, private scores or latency is treated as an ability measurement. Model and runtime documentation reviewed 2026-10-31; no fresh hardware execution.

Model information sources

Release date ↗ · Weights licence ↗ · Parameter count ↗

HF safetensors metadata for full weights. Memory estimate: BF16 weights plus 20% runtime allowance and 2 GB; not a measured requirement.

58.8±0.8#119 of 355 decision modelsfrom 96 benchmark results across 6 sources

$ how it compares

The decision models ranked either side of Clef-omni, out of 357 with a score, then more from Cloudflare.

#	model	score	where	RAM	released
119	gpt-5.6-luna (DecisionBench openrout) OpenAI	59±2	comparison only		2026-07-09
120	simplejev-qwen3.8-27b (DecisionBench 1d4bf0f2) featherless-ai	59±2	model files
121	Clef-omni Cloudflare	59±1	api $0.15/M input tokens +local	~87GB	2026-10
122	Decider chat · Qwen3.6-27B (decision readout) Mapika	59±1	source code
123	Xor 1.2 Juspay Technologies	59±1	model files		2026-09-28
	more from Cloudflare
74	Clef Cloudflare	63±1	api $0.24/M input tokens +local	~68GB	2026-10
134	Clef-flash Cloudflare	58±1	api $0.038/M input tokens +local	~25GB	2026-10

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

$ what it's good at

tools & automation▁▃▁▁▁▂▃▄▆██▆ #2 of 250 · 2.7% of its score · 6 tests

Beats Jeffan-31b-1710, baselabs mammoth-1 (27B), Sieve 31B v3 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 0.4.1: API-Bank (accuracy)	87.0% ✎	▁▁▁▁▁▁▁▂▃▂█▂ beats 211/236	easy · avg 70%	▇ 0.8%	sharp · noisy
Decision Index 0.4.1: BFCL (case exact accuracy)	98.6% ✎	▁▂▁▁▁▂▂▄▅▆█▃ beats 245/246	very easy · avg 92%	▇ 0.6%	sharp · maxed out · noisy
Decision Index 0.4.1: Home appliance simulator (case exact accuracy)	68.2% ✎	█▁▁▃▄▄▆▄▁▂▁▃ beats 168/243	medium · avg 37%	▇ 0.5%	sharp · at its level · noisy
Decision Index 0.4.1: When2Call (accuracy)	56.6% ✎	▁▁▁▂▃▅▅▆█▇█▃ beats 58/246	easy · avg 72%	▇ 0.4%	noisy
Decision Index 0.4.1: ToolRet (nDCG@10)	65.2% ✎	▁▁▁▁▁·▁▁▁▄█▁ beats 142/246	medium · avg 60%	▇ 0.4%	blunt · at its level
Decision Index 0.4.1: RouterBench (selected quality (quality objective))	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▁▁▁·▁▁▁▁▃▃█▅ #62 of 258 · 9.3% of its score · 15 tests

Beats AutoTrust-JEV-27B v1.1, oenogarum (Gemma 4 31B), Aleria Judge 1.1.1 (Gemma 4 31B) (decision readout) 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)	86.5% ✎	▂···▁··▃▄▂█▇ beats 10/29	easy · avg 84%	▇ 1.6%	typical
System One Bench: MASSIVE intent (Arabic)	91.5% ✎	▄·▁·▁▃▃▃█▄▇▇ beats 29/29	easy · avg 76%	▇ 1.4%	typical
LeanMCP · scifact rank noul	71.6% ✎	▂····▂···▃▅█ beats 2/11	medium · avg 62%	▇ 1.2%	at its level
Decision Index 0.4.1: BRIGHT (nDCG@10)	44.2% ✎	▁▁▁▁▁▁▁▂▂▄▇█ beats 141/246	medium · avg 38%	▇ 1.0%	blunt · at its level
Decision Index 0.4.1: Amazon ESCI (macro-F1)	57.8% ✎	▁▁▁▁▁▁▂▂▃▃█▅ beats 189/246	medium · avg 47%	▇ 0.9%	at its level
System One Bench: MIRACL rerank (Arabic)	83.0% ✎	▁▂·▁▁·▂·▄▄█▇ beats 9/29	easy · avg 79%	▇ 0.9%	typical
LeanMCP · nfcorpus rank noul	36.1% ✎	▃▃·····▃▃▅██ beats 2/11	hard · avg 29%	▇ 0.9%	blunt · at its level
System One Bench: MASSIVE scenario (Arabic)	89.0% ✎	▃▁·▄▃▂█▂▁▁·▃ beats 28/29	medium · avg 60%	▇ 0.6%	at its level

7 more tests

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: HoVer (accuracy)	62.5% ✎	▆▆▇▅▄▄▃▇▄█▆▄ beats 68/246	easy · avg 74%	▇ 0.2%	blunt
Decision Index 0.4.1: PhishNChips phishing decisions (accuracy)	72.5% ✎	█▅▅▄▃▂▁▁▁▁·▁ beats 140/245	easy · avg 70%	▇ 0.2%	noisy
System One Bench: AG News (English)	89.0% ✎	▁···▅▁▅▂▆█▂▅ beats 15/29	easy · avg 88%	▇ 0.2%	blunt
Decision Index 0.4.1: CLINC150+OOS (macro-F1)	97.5% ✎	▁▁▁▁▁▁▁▂▃▆█▄ beats 237/238	easy · avg 83%	▇ 0.1%	sharp · noisy
Decision Index 0.4.1: BANKING77 (macro-F1)	94.2% ✎	▁·▁▁▁▁▁▁▂█▇█ beats 238/238	easy · avg 79%	▇ 0.1%	noisy
System One Bench: Banking77 (English)	96.0% ✎	█▅▆█▅▃▃▃···▂ beats 25/26	very easy · avg 87%	▇ 0.0%	blunt · noisy
Decision Index 0.4.1: SGD/SGD-X (macro-F1)	43.1% ✎	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▂▂▂▃▂▅▆▃█▆▄▃ #7 of 256 · 13% of its score · 18 tests

Beats Jeffan-31b-1710, Sieve 31B v3, Fastino GLiDE 1.2 here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: ANLI (macro-F1)	63.6% ✎	▁▂▃▅▅█▂····▁ beats 118/246	medium · avg 60%	▇ 2.1%	at its level
Decision Index 0.4.1: NLI4CT (macro-F1)	77.9% ✎	▁▁▁▂▂▃▃▄▇▇█▄ beats 127/246	easy · avg 74%	▇ 2.0%	typical
Decision Index 0.4.1: HellaSwag (accuracy)	99.0% ✎	▂▄▂▂▄▇▃▇█▅▃▃ beats 246/246	very easy · avg 85%	▇ 1.7%	sharp · maxed out
System One Bench: BoolQ (English)	90.0% ✎	▁▁▁▃·▁▃▃▇▇█▅ beats 17/29	easy · avg 86%	▇ 1.1%	typical
System One Bench: OSACT4 offensive (Arabic)	90.5% ✎	▁······▁▃█▇▄ beats 18/29	easy · avg 84%	▇ 1.0%	sharp
Decision Index 0.4.1: iSarcasmEval (Sarcasm F1 · track A, English)	37.3% ✎	▁·▁·▁▁▂▄▅█▇▄ beats 80/246	medium · avg 44%	▇ 1.0%	at its level
System One Bench: SST-2 (English)	97.0% ✎	▁·▃▁▂▂▃▅█·▃▁ beats 27/29	easy · avg 92%	▇ 0.7%	typical
System One Bench: SST-5 (English)	51.0% ✎	▃·▁▁▃·▇▄▄▄█▅ beats 7/29	medium · avg 51%	▇ 0.5%	blunt · at its level

10 more tests

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: ACOS (per-review F1)	24.2% ✎	▁··▂▃▃▅▇█▇▃▃ beats 138/246	hard · avg 22%	▇ 0.5%	at its level · noisy
Decision Index 0.4.1: ContractNLI (macro-F1)	82.4% ✎	▁▁▁▁▁▁▁▂▂▇▇█ beats 185/246	easy · avg 71%	▇ 0.4%	noisy
Decision Index 0.4.1: VAST (macro-F1)	51.4% ✎	▁▁▁▂▅▄▄▂▄▆█▄ beats 80/246	medium · avg 62%	▇ 0.4%	at its level
Decision Index 0.4.1: RAGTruth (F1 on hallucinated class)	34.5% ✎	▁▁▁▁▁▁▁▃▁▃▆█ beats 32/246	easy · avg 66%	▇ 0.3%	at its level · noisy
System One Bench: XNLI Arabic (Arabic)	73.3% ✎	▄▁▃▁▄▇█▅▃▁·▃ beats 16/29	medium · avg 68%	▇ 0.3%	at its level
System One Bench: XNLI French (French)	82.7% ✎	▂▅·▁▁▅▅█▃▁·▃ beats 20/29	medium · avg 76%	▇ 0.3%	blunt · at its level
Decision Index 0.4.1: FinEntity (macro-F1)	96.7% ✎	▁···▁▁▁▁▃▆█▃ beats 245/246	easy · avg 84%	▇ 0.2%	noisy
System One Bench: Tweet sentiment (Arabic)	50.0% ✎	▁······▁▃▄█▄ beats 2/29	medium · avg 62%	▇ 0.2%	blunt · at its level
System One Bench: XNLI English (English)	82.7% ✎	▃·▁·▃▇█▄█▃▃▁ beats 5/29	easy · avg 85%	▇ 0.1%	blunt · noisy
System One Bench: DAIR Emotion (English)	56.0% ✎	██▃·····▁··▁ beats 9/29	medium · avg 59%	▇ 0.0%	blunt · at its level

arts & human taste▁▇▄▃▅▅█▅▄▂▁▁ #22 of 247 · 3.7% of its score · 6 tests

Beats Jeffan-31b-1710, vLLM-SR d3-turbo v1.1 (27B), WeMM-JEV-27B here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: New Yorker (accuracy)	66.5% ✎	▁▁▂▃▃▃▇▅█▅▃▁ beats 137/246	medium · avg 61%	▇ 1.5%	at its level
Decision Index 0.4.1: Humicroedit (accuracy)	71.1% ✎	▁·▁▃▂▂▅█▄▂▃▁ beats 244/246	medium · avg 60%	▇ 0.8%	blunt · at its level
Decision Index 0.4.1: cfcolor (accuracy)	63.2% ✎	▁········▁▇█ beats 151/246	medium · avg 60%	▇ 0.4%	blunt · at its level
Decision Index 0.4.1: BPoMP (accuracy)	97.3% ✎	▁█▄▄▅█▆▅▄▁▁▁ beats 226/246	easy · avg 83%	▇ 0.4%	sharp · noisy
Decision Index 0.4.1: POP909-CL (accuracy)	2.2% ✎	▄▅▅▃▆▅▃▅▂▅▃█ beats 47/237	hard · avg 21%	▇ 0.4%	sharp · at its level · noisy
Decision Index 0.4.1: Habermas Machine (accuracy)	70.0% ✎	▁▂▄█▇▅▁▁▄▄▁▃ beats 245/246	medium · avg 47%	▇ 0.1%	blunt · at its level

knowledge & reasoning▁▂▂▂▂▃▃▅▅▆█▂ #28 of 263 · 36% of its score · 21 tests

Beats Bespoke Nimble v4 Max (27B), vera-large, KnowLine 26B-A4B Gen2.5 here, though they rank higher overall. Trails Solar Decide here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Etsabary · 1342 text decisions	64.1% ✎	▁··▂·▃▅▁▃▂▂█ beats 4/24	medium · avg 47%	▇▇▇ 7.2%	sharp · at its level
Decision Index 0.4.1: MMLU-Pro (accuracy)	69.3% ✎	▄▄▃▃▂▄▅▇██▅▁ beats 204/246	medium · avg 46%	▇▇ 4.2%	sharp · at its level
LeanMCP · scienceqa text	92.1% ✎	▄··▄█▄█▄·█▄▄ beats 2/11	very easy · avg 88%	▇ 2.9%	sharp · maxed out
Decision Index 0.4.1: MMLU (accuracy)	92.3% ✎	▁▄▄▃▃▂▅▇▅█▇▃ beats 246/246	easy · avg 70%	▇ 2.6%	sharp
LeanMCP · medqa usmle	79.5% ✎	▄▄▄·▄▄▄▄▄·██ beats 3/11	easy · avg 72%	▇ 2.5%	typical
Decision Index 0.4.1: BBH (accuracy)	69.0% ✎	▄▆▄▄█▇▇▄▃▁▁▁ beats 125/246	easy · avg 66%	▇ 2.4%	typical
MedQA · Freyer et al. (first-valid accuracy)	79.4% ✎	█··▃▃·▃▃▃·▃█ beats 3/11	easy · avg 72%	▇ 1.9%	typical
LeanMCP · pubmedqa	67.4% ✎	▂···▂·▂·▄▄▂█ beats 5/11	medium · avg 54%	▇ 1.6%	at its level

13 more tests

test	result	vs the others in its run	difficulty	counts	why
LeanMCP · medmcqa	72.5% ✎	▃·▃··▃█▃·▅▅▃ beats 4/11	medium · avg 63%	▇ 1.6%	at its level
Decision Index 0.4.1: GSM8K (accuracy (4- and 10-choice))	67.8% ✎	▁··▄█▆▆▆▅▅▃▁ beats 181/246	medium · avg 45%	▇ 1.4%	sharp · at its level
Decision Index 0.4.1: GPQA Diamond (accuracy)	46.9% ✎	▂▆▅▅▆█▆▃▁··▁ beats 169/246	medium · avg 39%	▇ 1.3%	blunt · at its level
LeanMCP · mmlu pro	71.5% ✎	▄··▄▄████··▄ beats 9/11	medium · avg 54%	▇ 1.2%	at its level
Decision Index 0.4.1: ARC-Challenge (accuracy)	97.9% ✎	▂▂▂▂▂▂▁▃▄▆█▆ beats 241/246	very easy · avg 90%	▇ 1.2%	sharp · maxed out
Decision Index 0.4.1: WinoGrande (accuracy)	96.8% ✎	█▅█▅▄█▆▄▃▃▁▃ beats 244/246	easy · avg 78%	▇ 1.2%	typical
Decision Index 0.4.1: SATA-Bench (case exact accuracy)	31.3% ✎	▁▁▁▁▁▁▁▂▅█▂▁ beats 164/246	hard · avg 22%	▇ 0.6%	at its level
Decision Index 0.4.1: ARC-Easy (accuracy)	99.5% ✎	▁▁▂▁▂▂▂▂▃▆█▅ beats 245/246	very easy · avg 96%	▇ 0.6%	sharp · maxed out · noisy
Decision Index 0.4.1: MuSR (accuracy)	84.3% ✎	▁▁▂█▅▄▁▁▁▃▁▁ beats 238/246	medium · avg 60%	▇ 0.5%	blunt · at its level
Decision Index 0.4.1: CRUXEval (accuracy)	85.4% ✎	██▄▃▃▄▁▁···▁ beats 216/246	medium · avg 58%	▇ 0.4%	at its level · noisy
Decision Index 0.4.1: CLadder (accuracy)	97.2% ✎	▁███▆▅▃▃▁▁▁▃ beats 245/246	easy · avg 71%	▇ 0.4%	typical
Decision Index 0.4.1: ChessBench (accuracy)	21.9% ✎	▄▇▄▄▄▆▄█▄▁▁▁ beats 187/246	hard · avg 16%	▇ 0.3%	blunt
Decision Index 0.4.1: HLE (accuracy)	9.2% ✎	▁▁▃▃▆█▇▅▅▃▁▁ beats 15/246	very hard · avg 13%	▇ 0.0%	blunt

vision▁·▃▂▃▃▇▄▄█▄▃ #48 of 79 · 18% of its score · 18 tests

Trails LiquidAI d1-3B, JPT-9B Qwen3.5-9B LoRA, Solomon v1.1 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Etsabary · 200 image decisions	52.5% ✎	▂▃··▂▃▂·█▃▂▂ beats 1/14	medium · avg 45%	▇▇ 4.8%	sharp · at its level
System One Bench: Image object count	68.0% ✎	▃····▃█▃▅·█▃ beats 0/11	easy · avg 71%	▇ 2.4%	sharp
System One Bench: SST-2 screenshot	89.5% ✎	█···███·█·▄▄ beats 1/11	easy · avg 87%	▇ 1.8%	sharp
System One Bench: Visual yes/no questions	87.0% ✎	▂······▃█▃·▃ beats 5/11	medium · avg 80%	▇ 1.5%	at its level
System One Bench: Oxford pets classification	88.0% ✎	███·█▄▄·▄··▄ beats 8/11	easy · avg 77%	▇ 1.3%	typical
System One Bench: Arabic sentiment screenshot	28.5% ✎	▄··▄··▂···▆█ beats 0/11	very hard · avg 37%	▇ 0.8%	sharp · noisy
Decision Index Vision 0.4 · Mind2Web (chance-corrected points) submitted	74.7900 points ✎	▁·▁▁▃▁▄▄▃▅█▂ beats 29/61	avg 63.423 points	▇ 0.6%	blunt
Decision Index Vision 0.4 · MMMU-Pro vision (chance-corrected points) submitted	22.3700 points ✎	▁▁▅▁▄▃▄█▄▃▇▇ beats 15/61	avg 22.797 points	▇ 0.6%	blunt

10 more tests

test	result	vs the others in its run	difficulty	counts	why
Decision Index Vision 0.4 · Winoground (chance-corrected points) submitted	75.0000 points ✎	▁·▁▁▁▁·▂▁▃▅█ beats 18/61	avg 66.987 points	▇ 0.5%	blunt
Decision Index Vision 0.4: private same skills (points)	55.4000 points ✎	▁▁▃▃▁▂▄▆▃▃▅█ beats 31/61	avg 48.067 points	▇ 0.5%	blunt
Decision Index Vision 0.4 · Moderation (Hateful Memes) (chance-corrected points) submitted	31.2400 points ✎	▆▃▃▆▄▄▇██▁▆▆ beats 22/61	avg 28.443 points	▇ 0.5%	blunt
Decision Index Vision 0.4 · RealWorldQA (chance-corrected points) submitted	61.9900 points ✎	▁·▂▂▁▃▃▅▅▃▂█ beats 33/61	avg 53.651 points	▇ 0.4%	blunt
Decision Index Vision 0.4 · CharXiv (chance-corrected points) submitted	61.1500 points ✎	▂▁▁▁▁▃▃▅▅▆██ beats 14/61	avg 62.137 points	▇ 0.4%	blunt
Decision Index Vision 0.4 · InfographicVQA (chance-corrected points) submitted	87.4300 points ✎	▁·▁▂▁▁▃▁▃▂▅█ beats 23/61	avg 83.859 points	▇ 0.3%	blunt
Decision Index Vision 0.4 · R-Bench-M (chance-corrected points) submitted	25.6400 points ✎	▂▄▇▆█▄▄▇▂··▁ beats 36/61	avg 19.676 points	▇ 0.3%	blunt
Decision Index Vision 0.4 · BLINK (chance-corrected points) submitted	48.4500 points ✎	▁···▁▂▂▃█▄▂▃ beats 16/61	avg 44.994 points	▇ 0.3%	blunt
Decision Index Vision 0.4 · CV-Bench (chance-corrected points) submitted	73.3700 points ✎	▁▁▁▁▂▁▁▁▄█▃▃ beats 24/61	avg 67.662 points	▇ 0.2%	blunt
Decision Index Vision 0.4 · KIE (CORD+FUNSD) (chance-corrected points) submitted	89.0500 points ✎	▁··▁▁▁▁▁▂▄█▁ beats 18/61	avg 85.184 points	▇ 0.2%	blunt

workflows▁▁▁▂▁▂▂▆█▄▁▁ #71 of 114 · 14% of its score · 12 tests

Trails Solar Decide Flash, InternLM Intern-Decision Qwen3.5-4B, StartLux-Decision v1.1 0.8B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LeanMCP · mcpbench server routing	85.7% ✎	█·███··▄···▄ beats 0/9	easy · avg 82%	▇ 3.0%	sharp
LeanMCP · mcpbench tool selection	92.3% ✎	▃··········█ beats 0/9	very easy · avg 89%	▇ 1.8%	sharp · maxed out · noisy
LeanMCP · mcpmark server routing	46.7% ✎	▂··▂··▄█▂·▂▄ beats 3/11	medium · avg 36%	▇ 1.4%	at its level
LeanMCP · rewardbench2	56.0% ✎	▄▄▄·▄▄▄▄█·█▄ beats 0/11	medium · avg 58%	▇ 1.4%	at its level
System One Bench: Typed decisions (English)	67.0% ✎	▁▁·▁▁▁▅▄▇██▁ beats 10/29	medium · avg 63%	▇ 1.3%	blunt · at its level
LeanMCP · aegis2 response	79.0% ✎	█··▃·▃·█··▅▅ beats 6/11	easy · avg 70%	▇ 1.2%	typical
LeanMCP · mcpmark tool selection	100.0% ✎	▂··········█ beats 7/11	very easy · avg 96%	▇ 1.0%	sharp · maxed out · noisy
LeanMCP · atbench500	67.4% ✎	▃··▅▅█··▅··▅ beats 1/11	medium · avg 63%	▇ 0.9%	sharp · at its level · noisy

4 more tests

test	result	vs the others in its run	difficulty	counts	why
LeanMCP · helpsteer2 helpfulness	1.1024 ordinal levels ✎	▄▄▄▄▄·▄·█▄▄▄ beats 1/10	avg 1.357 ordinal levels	▇ 0.9%	blunt
LeanMCP · sst5	0.5233 ordinal levels ✎	▂·····▂▂▄█▄▂ beats 8/11	avg 0.879 ordinal levels	▇ 0.8%	blunt
LeanMCP · prompt injections	71.6% ✎	▂▂▂▂█·▄▂···▂ beats 4/11	medium · avg 62%	▇ 0.6%	at its level
LeanMCP · jailbreak classification	92.5% ✎	▂·▄··▂···█▆▂ beats 2/11	very easy · avg 95%	▇ 0.0%	blunt · maxed out · noisy

other tests2 not in a task · 3.4% of its score

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: private same skills (equated points)	45.9900 points ✎	▂▂▂▃▃▃▄▅▅▇█▃ beats 125/247	avg 41.206 points	▇ 2.2%	blunt
Decision Index 0.4.1: private new domains (equated points)	43.6100 points ✎	▃█▃▃▂▆▃▃▄▄▄▄ beats 138/247	avg 38.861 points	▇ 1.2%	blunt

$ where to run it

Through an API

api · $0.15/M input tokens · developers.cloudflare.com ↗ · price sourceCloudflare Workers AI list price, observed 2026-10-31. USD per million input tokens, including tokenized media; no output-token charge. Not a fixed price per decision.

Other ways to run it

Run it as published · needs ~86.6 GB, NVIDIA only · from huggingface.co/Cloudflare (official)
pip install torch==2.11 transformers==5.10.2 huggingface-hub pillow av && python -c 'import sys; from huggingface_hub import snapshot_download; p=snapshot_download("Cloudflare/clef-omni",revision="f32dfc060386af3eadec676c64b34f24ece2bcde"); sys.path.insert(0,p); from joint_schema_model import load_release_model,systemone; m,t=load_release_model(p,device="cuda"); print(systemone(m,t,{"model":"clef-omni","state":"The invoice is overdue.","questions":{"overdue":{"type":"noul","instructions":"Is the invoice overdue?"}}}))'

Download model files ↗
Published model files verified. This is a verified repository snapshot; the benchmark does not establish an identical weight revision. See the local setup below for the reviewed runtime. Source · checked 2026-10-31.

Model repository and documentation: huggingface.co/Cloudflare ↗

$ how the score was worked out · Decision Index 0.4.1 38%, LeanMCP · JevBench 25%, System One Bench (Ali Alhashi… 18% and 3 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	42	▇▇▇▇▇▇▇▇▇▇▇ 38%	CC BY 4.0
LeanMCP · JevBench	18	▇▇▇▇▇▇▇ 25%	MIT
System One Bench (Ali Alhashimi)	21	▇▇▇▇▇ 18%	CC-BY-4.0
Etsabary · decision benchmarks	2	▇▇▇▇ 12%	MIT
Decision Index Vision 0.4 · submitted; independent execution unverified	12	▇▇ 5%	CC BY 4.0
Freyer et al. · MedQA	1	▇ 2%	CC BY 4.0
prior · pull toward the middle		▇ 0%
weighted average		= 58.8

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]	43.1%	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