Clef-flash

+ contribute

Clef-flash is an open-weights decision model from Cloudflare, released October 2026. Its published size is 9.41B parameters. Licence: apache-2.0.

Catalogue identity notes

Exact Decision Index 0.4.1 text-board engine clef-flash. 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.

57.6±0.6#132 of 355 decision modelsfrom 168 benchmark results across 12 sources

$ how it compares

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

#	model	score	where	RAM	released
132	deepseek-v4.1-flash (DecisionBench openrout) DeepSeek	58±2	model files
133	Nace.AI Drex v1.5 Nace.AI	58±1	model files
134	Clef-flash Cloudflare	58±1	api $0.038/M input tokens +local	~25GB	2026-10
135	Bespoke Nimble 9B v3 Bespoke Labs	58±1	local	~21GB	2026-10-06
136	simplejev-qwen3.6-35b-a3b (DecisionBench 995ad96e) featherless-ai	58±2	model files
	more from Cloudflare
74	Clef Cloudflare	63±1	api $0.24/M input tokens +local	~68GB	2026-10
121	Clef-omni Cloudflare	59±1	api $0.15/M input tokens +local	~87GB	2026-10

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

$ what it's good at

tools & automation▁▃▁▁▁▂▃▄▆██▆ #1 of 250 · 3.6% of its score · 10 tests

Beats Jeffan-31b-1710, baselabs mammoth-1 (27B), Sieve 31B v3 here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
DecideBench v1: agent routing	92.0% ✎	▄·····▃▄·█·▄ beats 2/10	very easy · avg 93%	▇ 0.6%	sharp · maxed out · noisy
AgentDecisionBench: L1 BFCL v4 + When2Call (3,991 items)	81.4% ✎	█·····▄·▄▄·▄ beats 4/5	easy · avg 75%	▇ 0.6%	typical
AgentDecisionBench: L2 DecisionGym common states (200 tasks)	47.7% ✎	██···█·██··█ beats 2/5	medium · avg 57%	▇ 0.5%	sharp · at its level · noisy
Decision Index 0.4.1: API-Bank (accuracy) submitted	93.1% ✎	▁▁▁▁▁▁▁▂▃▂█▃ beats 236/236	easy · avg 70%	▇ 0.5%	sharp · noisy
DecideBench v1: action review	88.0% ✎	▃▃·▃·█···█·▅ beats 4/10	easy · avg 87%	▇ 0.4%	sharp · noisy
Decision Index 0.4.1: BFCL (case exact accuracy) submitted	98.8% ✎	▁▂▁▁▁▂▂▄▅▆█▃ beats 246/246	very easy · avg 92%	▇ 0.3%	sharp · maxed out · noisy
Decision Index 0.4.1: Home appliance simulator (case exact accuracy) submitted	97.7% ✎	█▁▁▃▄▄▆▄▁▂▃▃ beats 220/243	medium · avg 37%	▇ 0.3%	sharp · at its level · noisy
Decision Index 0.4.1: When2Call (accuracy) submitted	65.3% ✎	▁▁▁▂▃▅▅▆█▇█▃ beats 82/246	easy · avg 72%	▇ 0.2%	noisy

2 more tests

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: ToolRet (nDCG@10) submitted	66.2% ✎	▁▁▁▁▁·▁▁▁▄█▁ beats 167/246	medium · avg 60%	▇ 0.2%	blunt · at its level
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▁▁▁·▁▁▁▁▃▃█▅ #68 of 258 · 8.9% of its score · 22 tests

Beats AutoTrust-JEV-27B v1.1, oenogarum (Gemma 4 31B), Decider 31B (NVFP4) 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
DecisionPivot v1.1 · paws (accuracy)	79.1% ✎	█·██·······█ beats 0/3	easy · avg 68%	▇ 1.3%	sharp
DecisionPivot v1.1 · nevir (pair_accuracy)	61.6% ✎	█··█··█····█ beats 0/3	medium · avg 49%	▇ 1.0%	sharp · at its level
System One Bench: MIRACL pairwise (Arabic)	90.0% ✎	▂···▁··▃▄▂█▇ beats 20/29	easy · avg 84%	▇ 0.9%	at its level
System One Bench: MASSIVE intent (Arabic)	90.5% ✎	▄·▁·▁▃▃▃█▄▇▇ beats 26/29	easy · avg 76%	▇ 0.8%	typical
LeanMCP · scifact rank noul	71.7% ✎	▂····▂···▃▅█ beats 3/11	medium · avg 62%	▇ 0.6%	at its level
Decision Index 0.4.1: BRIGHT (nDCG@10) submitted	38.3% ✎	▁▁▁▁▁▁▁▂▂▄▇█ beats 86/246	medium · avg 38%	▇ 0.5%	blunt · at its level
System One Bench: MIRACL rerank (Arabic)	88.0% ✎	▁▂·▁▁·▂·▄▄█▇ beats 19/29	easy · avg 79%	▇ 0.5%	typical
Decision Index 0.4.1: Amazon ESCI (macro-F1) submitted	57.4% ✎	▁▁▁▁▁▁▂▂▃▃█▅ beats 187/246	medium · avg 47%	▇ 0.5%	at its level

14 more tests

test	result	vs the others in its run	difficulty	counts	why
LeanMCP · nfcorpus rank noul	36.9% ✎	▃▃·····▃▃▅██ beats 5/11	hard · avg 29%	▇ 0.5%	blunt · at its level
DecisionPivot v1.1 · entity match amazon google (auroc)	0.9510 auroc ✎	█···█·█····█ beats 0/3	avg 0.290 auroc	▇ 0.4%	typical
DecisionPivot v1.1 · entity match abt buy (auroc)	0.9790 auroc ✎	█······█··██ beats 0/3	avg 0.320 auroc	▇ 0.4%	typical
System One Bench: MASSIVE scenario (Arabic)	89.5% ✎	▃▁·▄▃▂█▂▁▁·▃ beats 29/29	medium · avg 60%	▇ 0.3%	at its level
DecisionPivot v1.1 · rerank nfcorpus (ndcg@10)	0.4260 ndcg@10 ✎	█··········▄ beats 1/2	avg -0.224 ndcg@10	▇ 0.3%	small run
DecisionPivot v1.1 · rerank scifact (ndcg@10)	0.8150 ndcg@10 ✎	█·········██ beats 2/2	avg 0.162 ndcg@10	▇ 0.3%	small run
DecideBench v1: support intent	94.0% ✎	▃····▂▃·▃··█ beats 3/10	very easy · avg 95%	▇ 0.2%	sharp · maxed out · noisy
Decision Index 0.4.1: HoVer (accuracy) submitted	61.2% ✎	▆▆▇▅▄▄▃▇▄█▆▄ beats 57/246	easy · avg 74%	▇ 0.1%	blunt
Decision Index 0.4.1: PhishNChips phishing decisions (accuracy) submitted	75.0% ✎	█▅▅▄▃▂▁▁▁▁·▁ beats 152/245	easy · avg 70%	▇ 0.1%	noisy
System One Bench: AG News (English)	90.0% ✎	▁···▅▁▅▂▆█▂▅ beats 19/29	easy · avg 88%	▇ 0.1%	blunt
Decision Index 0.4.1: CLINC150+OOS (macro-F1) submitted	66.9% ✎	▁▁▁▁▁▁▁▃▃▆█▄ beats 40/238	easy · avg 83%	▇ 0.1%	sharp · noisy
Decision Index 0.4.1: BANKING77 (macro-F1) submitted	90.8% ✎	▁·▁▁▁▁▁▁▂█▇█ beats 194/238	easy · avg 79%	▇ 0.1%	noisy
System One Bench: Banking77 (English)	98.5% ✎	█▅▆█▅▃▃▃···▃ beats 26/26	very easy · avg 87%	▇ 0.0%	blunt · noisy
Decision Index 0.4.1: SGD/SGD-X (macro-F1) submitted	34.4% ✎	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▂▂▂▃▂▅▆▃█▆▄▃ #18 of 256 · 8.9% of its score · 21 tests

Beats KnowLine 31B Gen1, Ainvest-Decision-27B v1.0, 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: ANLI (macro-F1) submitted	59.1% ✎	▁▂▃▅▅█▂····▁ beats 103/246	medium · avg 60%	▇ 1.1%	at its level
Decision Index 0.4.1: NLI4CT (macro-F1) submitted	78.6% ✎	▁▁▁▂▂▃▃▄▇▇█▄ beats 139/246	easy · avg 74%	▇ 1.1%	typical
DecideBench v1: review sentiment	92.0% ✎	▂··▂···▂▂█▄▂ beats 6/10	easy · avg 85%	▇ 1.0%	sharp
Decision Index 0.4.1: HellaSwag (accuracy) submitted	98.6% ✎	▂▄▂▂▄▇▃▇█▅▃▃ beats 245/246	very easy · avg 85%	▇ 1.0%	sharp
DecideBench v1: content moderation	86.0% ✎	▂··▂···▄▄▂·█ beats 4/10	easy · avg 83%	▇ 0.6%	sharp · noisy
System One Bench: BoolQ (English)	88.5% ✎	▁▁▁▃·▁▃▃▇▇█▅ beats 11/29	easy · avg 86%	▇ 0.6%	typical
System One Bench: OSACT4 offensive (Arabic)	89.5% ✎	▁······▁▃█▇▄ beats 13/29	easy · avg 84%	▇ 0.6%	sharp
Decision Index 0.4.1: iSarcasmEval (Sarcasm F1 · track A, English) submitted	19.7% ✎	▁·▁·▁▁▂▄▅█▇▄ beats 23/246	medium · avg 44%	▇ 0.5%	at its level

13 more tests

test	result	vs the others in its run	difficulty	counts	why
System One Bench: SST-2 (English)	95.0% ✎	▁·▃▁▂▂▃▅█·▂▁ beats 16/29	easy · avg 92%	▇ 0.4%	typical
System One Bench: SST-5 (English)	57.0% ✎	▃·▁▁▃·▇▄▄▄█▅ beats 18/29	medium · avg 51%	▇ 0.3%	blunt · at its level
DecideBench v1: claim support	66.0% ✎	▄··▄·▄·▄▄███ beats 2/10	easy · avg 82%	▇ 0.3%	sharp · noisy
Decision Index 0.4.1: ACOS (per-review F1) submitted	26.0% ✎	▁··▂▃▃▅▇█▇▃▃ beats 147/246	hard · avg 22%	▇ 0.3%	at its level · noisy
Decision Index 0.4.1: ContractNLI (macro-F1) submitted	84.5% ✎	▁▁▁▁▁▁▁▂▂▇▇█ beats 220/246	easy · avg 71%	▇ 0.2%	noisy
Decision Index 0.4.1: VAST (macro-F1) submitted	49.6% ✎	▁▁▁▂▅▄▄▂▄▆█▄ beats 75/246	medium · avg 62%	▇ 0.2%	at its level
Decision Index 0.4.1: RAGTruth (F1 on hallucinated class) submitted	35.6% ✎	▁▁▁▁▁▁▁▃▁▃▆█ beats 33/246	easy · avg 66%	▇ 0.2%	at its level · noisy
System One Bench: XNLI Arabic (Arabic)	68.7% ✎	▄▁▃▁▄▇█▅▃▁·▃ beats 8/29	medium · avg 68%	▇ 0.2%	at its level
System One Bench: XNLI French (French)	80.0% ✎	▂▅·▁▁▅▅█▃▁·▃ beats 13/29	medium · avg 76%	▇ 0.1%	blunt · at its level
Decision Index 0.4.1: FinEntity (macro-F1) submitted	97.1% ✎	▁···▁▁▁▁▃▆█▃ beats 246/246	easy · avg 84%	▇ 0.1%	noisy
System One Bench: Tweet sentiment (Arabic)	47.0% ✎	▁······▃▁▄█▄ beats 1/29	medium · avg 62%	▇ 0.1%	blunt · at its level
System One Bench: XNLI English (English)	84.7% ✎	▃·▁·▃▇█▄█▃▃▁ beats 9/29	easy · avg 85%	▇ 0.0%	blunt · noisy
System One Bench: DAIR Emotion (English)	57.0% ✎	██▃·····▁··▁ beats 16/29	medium · avg 59%	▇ 0.0%	blunt · at its level

arts & human taste▁▇▄▃▅▅█▅▄▂▁▁ #44 of 247 · 2.0% of its score · 6 tests

Beats Jeffan-31b-1710, vLLM-SR d3-turbo v1.1 (27B), Nace.AI Drex Max v1 (27B) here, though they rank higher overall. Trails Kas 4B, 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: New Yorker (accuracy) submitted	66.3% ✎	▁▁▂▃▃▃▇▅█▅▃▁ beats 136/246	medium · avg 61%	▇ 0.8%	at its level
Decision Index 0.4.1: Humicroedit (accuracy) submitted	75.1% ✎	▁·▁▃▂▂▅█▄▂▁▃ beats 246/246	medium · avg 60%	▇ 0.4%	blunt · at its level
Decision Index 0.4.1: cfcolor (accuracy) submitted	65.9% ✎	▁········▁▇█ beats 204/246	medium · avg 60%	▇ 0.2%	blunt · at its level
Decision Index 0.4.1: BPoMP (accuracy) submitted	95.4% ✎	▁█▄▄▅█▆▅▄▁▁▁ beats 200/246	easy · avg 83%	▇ 0.2%	sharp · noisy
Decision Index 0.4.1: POP909-CL (accuracy) submitted	1.6% ✎	▄▅▅▃▆▅▃▅▂▅▃█ beats 35/237	hard · avg 21%	▇ 0.2%	sharp · at its level · noisy
Decision Index 0.4.1: Habermas Machine (accuracy) submitted	71.8% ✎	▁▂▄█▇▅▁▁▄▄▁▃ beats 246/246	medium · avg 47%	▇ 0.1%	blunt · at its level

knowledge & reasoning▁▂▂▂▂▃▃▅▅▆█▂ #47 of 263 · 24% of its score · 30 tests

Beats OneJev1.6-27B, baselabs elephant-1 (9B), StartLux-Decision v1.1 9B 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 5/24	medium · avg 47%	▇▇ 4.0%	sharp · at its level
Decision Index 0.4.1: MMLU-Pro (accuracy) submitted	65.2% ✎	▄▄▃▃▂▄▅▇██▅▁ beats 182/246	medium · avg 46%	▇ 2.3%	sharp · at its level
LeanMCP · scienceqa text	92.3% ✎	▄··▄█▄█▄·█▄▄ beats 3/11	very easy · avg 88%	▇ 1.6%	sharp · maxed out
Decision Index 0.4.1: MMLU (accuracy) submitted	91.9% ✎	▁▄▄▃▃▂▅▇▅█▇▃ beats 245/246	easy · avg 70%	▇ 1.4%	sharp
LeanMCP · medqa usmle	76.6% ✎	▄▄▄·▄▄▄▄▄·██ beats 2/11	easy · avg 72%	▇ 1.4%	typical
Decision Index 0.4.1: BBH (accuracy) submitted	68.9% ✎	▄▆▄▄█▇▇▄▃▁▁▁ beats 124/246	easy · avg 66%	▇ 1.3%	at its level
MedQA · Freyer et al. (first-valid accuracy)	73.7% ✎	█··▃▃·▃▃▃·▃█ beats 2/11	easy · avg 72%	▇ 1.1%	typical
DecisionPivot v1.1 · folio validation (accuracy)	69.1% ✎	█·········▄▄ beats 0/3	medium · avg 56%	▇ 1.0%	sharp · at its level

22 more tests

test	result	vs the others in its run	difficulty	counts	why
LeanMCP · medmcqa	70.4% ✎	▃·▃··▃█▃·▅▅▃ beats 2/11	medium · avg 63%	▇ 0.9%	at its level
LeanMCP · pubmedqa	69.6% ✎	▂···▂·▂·▄▄▂█ beats 9/11	medium · avg 54%	▇ 0.9%	at its level
DecisionPivot v1.1 · cuad clause (accuracy)	74.7% ✎	▄·········█▄ beats 0/3	medium · avg 65%	▇ 0.8%	at its level
Decision Index 0.4.1: GSM8K (accuracy (4- and 10-choice)) submitted	46.7% ✎	▁··▄█▆▆▆▅▅▃▁ beats 133/246	medium · avg 45%	▇ 0.8%	sharp · at its level
Decision Index 0.4.1: GPQA Diamond (accuracy) submitted	51.5% ✎	▂▆▅▅▆█▆▃▁··▁ beats 209/246	medium · avg 39%	▇ 0.7%	blunt · at its level
Decision Index 0.4.1: ARC-Challenge (accuracy) submitted	98.2% ✎	▂▂▂▂▂▂▁▃▄▆█▆ beats 246/246	very easy · avg 90%	▇ 0.7%	sharp · maxed out
LeanMCP · mmlu pro	67.6% ✎	▄··▄▄████··▄ beats 6/11	medium · avg 54%	▇ 0.7%	at its level
Decision Index 0.4.1: WinoGrande (accuracy) submitted	97.5% ✎	█▅█▅▄█▆▄▃▃▁▃ beats 246/246	easy · avg 78%	▇ 0.6%	typical
DecisionPivot v1.1 · legalbench mix (accuracy)	73.9% ✎	█····█···█·█ beats 0/3	medium · avg 62%	▇ 0.6%	at its level
DecisionPivot v1.1 · primevul pair (pair_accuracy)	0.2% ✎	▄·········█▄ beats 0/3	very hard · avg 1.2%	▇ 0.5%	sharp · too hard for it
Ollaya · PubMedQA	77.6% ✎	▃▂▃▃·▂▃▃▆█▆▂ beats 24/25	easy · avg 69%	▇ 0.5%	typical
Ollaya · BoolQ	85.7% ✎	▁··▃▃▁▄▁▃█▇▄ beats 15/25	easy · avg 84%	▇ 0.3%	typical
Decision Index 0.4.1: ARC-Easy (accuracy) submitted	99.5% ✎	▁▁▂▁▂▂▂▂▃▆█▅ beats 246/246	very easy · avg 96%	▇ 0.3%	sharp · maxed out · noisy
Decision Index 0.4.1: SATA-Bench (case exact accuracy) submitted	36.6% ✎	▁▁▁▁▁▁▁▂▅█▂▁ beats 212/246	hard · avg 22%	▇ 0.3%	at its level
Decision Index 0.4.1: MuSR (accuracy) submitted	86.2% ✎	▁▁▂█▅▄▁▁▁▃▁▁ beats 240/246	medium · avg 60%	▇ 0.3%	blunt · at its level
Decision Index 0.4.1: CRUXEval (accuracy) submitted	86.5% ✎	██▄▃▃▄▁▁···▁ beats 225/246	medium · avg 58%	▇ 0.2%	at its level · noisy
Decision Index 0.4.1: CLadder (accuracy) submitted	97.7% ✎	▁███▆▅▃▃▁▁▁▃ beats 246/246	easy · avg 71%	▇ 0.2%	typical
Ollaya · SQuAD 2 answerability	82.3% ✎	▁▁▃▁▇█▃▃▄▃·▁ beats 19/25	easy · avg 80%	▇ 0.2%	blunt
Ollaya · VitaminC verification	76.8% ✎	▁···▃·▄█▃··▁ beats 13/25	easy · avg 78%	▇ 0.2%	blunt
Decision Index 0.4.1: ChessBench (accuracy) submitted	23.1% ✎	▄▇▄▄▄▆▄█▄▁▁▁ beats 203/246	hard · avg 16%	▇ 0.2%	blunt
Ollaya · MultiNLI	82.3% ✎	▂··▄██▂▆▆█▆▂ beats 7/25	very easy · avg 87%	▇ 0.1%	blunt
Decision Index 0.4.1: HLE (accuracy) submitted	11.8% ✎	▁▁▂▃▆█▇▅▅▃▁▁ beats 102/246	very hard · avg 13%	▇ 0.0%	blunt

vision▁·▃▂▃▃▇▄▄█▄▃ #14 of 79 · 7.5% of its score · 10 tests

Beats Fastino GLiDE 1.2, StartLux-Decision v1.1 27B, vLLM-SR d3-turbo v1.1 (27B) here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
Etsabary · 200 image decisions	63.0% ✎	▂▂··▂▃▂·█▃▂▂ beats 3/14	medium · avg 45%	▇ 2.6%	sharp · at its level
System One Bench: Image object count	88.0% ✎	▃····▃█▃▅·█▃ beats 3/11	easy · avg 71%	▇ 1.4%	sharp
System One Bench: SST-2 screenshot	95.5% ✎	█···███·█·▄▄ beats 8/11	easy · avg 87%	▇ 1.0%	sharp
System One Bench: Visual yes/no questions	88.0% ✎	▂······▃█▃·▃ beats 8/11	medium · avg 80%	▇ 0.8%	at its level
System One Bench: Oxford pets classification	85.0% ✎	███·█▄▄·▄··▄ beats 4/11	easy · avg 77%	▇ 0.7%	typical
System One Bench: Arabic sentiment screenshot	33.5% ✎	▄··▄··▂···▆█ beats 1/11	very hard · avg 37%	▇ 0.4%	sharp · noisy
JevalKit: VisualDecisionBench images	82.2% ✎	▄▄···█▄·█▄█▄ beats 6/10	easy · avg 78%	▇ 0.3%	blunt
Benchmark Heaven imagejevbench v0.3.0: intelligence (points)	37.2832 points ✎	▃·▃▃··▅█▅▅·▃ beats 4/12	avg 32.235 points	▇ 0.2%	blunt

2 more tests

test	result	vs the others in its run	difficulty	counts	why
Benchmark Heaven imagejevbench v0.3.0: calibration (points)	67.2009 points ✎	▃···▃·▅█▅▃▃▅ beats 2/12	avg 68.584 points	▇ 0.1%	blunt
JevalKit: VisualDecisionBench video	87.8% ✎	no other model yet		—	not counted yet: no other model in the same run to compare it with

workflows▁▁▁▂▁▂▂▆█▄▁▁ #66 of 114 · 42% of its score · 67 tests

Beats Clef-omni here, though they rank higher overall. Trails InternLM Intern-Decision Qwen3.5-4B, StartLux-Decision v1.1 0.8B, Kev 4B r10 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LeanMCP · mcpbench server routing	87.5% ✎	█·███··▄···▄ beats 1/9	easy · avg 82%	▇ 1.7%	sharp
DecisionPivot v1.1 · mind2web element (accuracy)	64.8% ✎	█···█·█····█ beats 0/3	medium · avg 54%	▇ 1.6%	sharp · at its level
JevalKit: Procedural Typed Decisions	67.9% ✎	█·███▄·▄▄▄·▄ beats 6/12	easy · avg 67%	▇ 1.3%	sharp
DecisionPivot v1.1 · judgebench (accuracy)	62.2% ✎	█······██··█ beats 0/3	medium · avg 52%	▇ 1.2%	sharp · at its level
DecisionPivot v1.1 · llmbar (accuracy)	71.2% ✎	██·····█···█ beats 0/3	medium · avg 58%	▇ 1.2%	sharp · at its level
DecisionPivot v1.1 · processbench (accuracy)	50.7% ✎	█·····█··█·█ beats 0/3	medium · avg 41%	▇ 1.2%	sharp · at its level
DecisionPivot v1.1 · rewardbench (accuracy)	86.7% ✎	█······█·█·█ beats 0/3	easy · avg 83%	▇ 1.1%	sharp
LeanMCP · mcpbench tool selection	92.3% ✎	▃··········█ beats 1/9	very easy · avg 89%	▇ 1.1%	sharp · maxed out · noisy

59 more tests

test	result	vs the others in its run	difficulty	counts	why
DecisionPivot v1.1 · squad2 answerable (accuracy)	80.4% ✎	███········█ beats 0/3	easy · avg 69%	▇ 1.1%	sharp
DecisionPivot v1.1 · rewardbench2 (accuracy)	67.2% ✎	█·····█···██ beats 0/3	medium · avg 57%	▇ 1.0%	sharp · at its level
Support routing · DecisionModelHub	87.0% ✎	▄······█···▄ beats 0/3	easy · avg 82%	▇ 1.0%	sharp
DecisionPivot v1.1 · browsesafe (accuracy)	65.7% ✎	▄·······▄··█ beats 0/3	medium · avg 56%	▇ 1.0%	sharp · at its level
DecideBench v1: ticket triage	90.0% ✎	▄·····▃█·▆·▂ beats 2/10	very easy · avg 90%	▇ 0.9%	sharp · maxed out
DecisionPivot v1.1 · answerable facts (pair accuracy)	23.7% ✎	█·█········█ beats 0/2	hard · avg 18%	▇ 0.9%	sharp · at its level · noisy · small run
DecisionPivot v1.1 · perm inherit (pair accuracy)	21.3% ✎	██·········█ beats 0/2	very hard · avg 15%	▇ 0.8%	sharp · at its level · small run
DecisionPivot v1.1 · state goal (pair accuracy)	11.3% ✎	█···█······█ beats 0/2	very hard · avg 11%	▇ 0.8%	sharp · noisy · small run
DecisionPivot v1.1 · toolcall valid (pair accuracy)	40.7% ✎	█··········▄ beats 0/2	hard · avg 27%	▇ 0.8%	sharp · at its level · small run
DecisionPivot v1.1 · triage order (pair accuracy)	8.5% ✎	█···█······█ beats 0/2	very hard · avg 6.4%	▇ 0.8%	sharp · small run
DecisionPivot v1.1 · sched conflict (pair accuracy)	5.0% ✎	█·····█····█ beats 0/2	very hard · avg 7.1%	▇ 0.8%	sharp · small run
LeanMCP · rewardbench2	60.0% ✎	▄▄▄·▄▄▄▄█·█▄ beats 1/11	medium · avg 58%	▇ 0.8%	at its level
LeanMCP · mcpmark server routing	47.6% ✎	▂··▂··▄█▂·▂▄ beats 4/11	medium · avg 36%	▇ 0.8%	at its level
DecisionPivot v1.1 · filter match (pair accuracy)	3.7% ✎	█·····█····█ beats 0/2	very hard · avg 4.9%	▇ 0.8%	sharp · small run
Ollaya · Typed decisions test	70.3% ✎	▄·▂▆▄▄▂·▄█▆▆ beats 18/22	medium · avg 64%	▇ 0.8%	at its level
Support urgency · DecisionModelHub	95.0% ✎	█···█·····██ beats 0/3	very easy · avg 94%	▇ 0.7%	sharp · maxed out · noisy
System One Bench: Typed decisions (English)	70.5% ✎	▁▁·▁▁▁▅▄▇██▁ beats 14/29	medium · avg 63%	▇ 0.7%	blunt · at its level
DecisionPivot v1.1 · ledger reconcile (pair accuracy)	13.4% ✎	█·······█··█ beats 0/2	very hard · avg 12%	▇ 0.7%	sharp · small run
DecisionPivot v1.1 · clause change (pair accuracy)	9.0% ✎	█·█········█ beats 0/2	very hard · avg 10.0%	▇ 0.7%	sharp · noisy · small run
LeanMCP · aegis2 response	79.0% ✎	█··▃·▃·█··▅▅ beats 7/11	easy · avg 70%	▇ 0.7%	typical
DecisionPivot v1.1 · record rule (pair accuracy)	0.7% ✎	█········█·█ beats 0/2	very hard · avg 6.0%	▇ 0.7%	sharp · noisy · small run
DecisionPivot v1.1 · policy case (pair accuracy)	19.9% ✎	██·········█ beats 0/2	very hard · avg 14%	▇ 0.7%	sharp · at its level · noisy · small run
LeanMCP · mcpmark tool selection	100.0% ✎	▂··········█ beats 8/11	very easy · avg 96%	▇ 0.6%	sharp · maxed out · noisy
DecisionPivot v1.1 · bipia injection (accuracy)	77.3% ✎	█····█····██ beats 0/3	easy · avg 76%	▇ 0.6%	sharp · noisy
JevalKit: DecisionBench standard eval	78.5% ✎	▃▃·▃···▅█▅·█ beats 9/12	easy · avg 70%	▇ 0.6%	typical
DecisionPivot v1.1 · log count check (pair accuracy)	0.0% ✎	█·····█····█ beats 0/2	very hard · avg 4.6%	▇ 0.6%	sharp · noisy · small run
DecisionPivot v1.1 · dep resolve (pair accuracy)	6.0% ✎	██·········█ beats 0/2	very hard · avg 6.8%	▇ 0.5%	sharp · noisy · small run
DecisionPivot v1.1 · known unknown (accuracy)	80.3% ✎	█··█······██ beats 1/3	easy · avg 69%	▇ 0.5%	typical
DecisionPivot v1.1 · cap boundary record (MSE skill)	-0.0100 skill ✎	█··█····█··█ beats 0/3	avg -0.616 skill	▇ 0.5%	typical
LeanMCP · atbench500	76.8% ✎	▃··▅▅█··▅··▅ beats 6/11	medium · avg 63%	▇ 0.5%	sharp · at its level · noisy
DecisionPivot v1.1 · agentreward success (accuracy)	73.9% ✎	█···██·····█ beats 2/3	medium · avg 58%	▇ 0.5%	at its level
DecisionPivot v1.1 · cap boundary card (MSE skill)	-0.0710 skill ✎	█··█····█··█ beats 0/3	avg -0.709 skill	▇ 0.5%	typical
DecisionPivot v1.1 · faithbench (accuracy)	51.3% ✎	█·█·█······█ beats 0/3	medium · avg 38%	▇ 0.5%	at its level
DecisionPivot v1.1 · aegis2 prompt (accuracy)	84.0% ✎	█·····█·█··█ beats 1/3	easy · avg 73%	▇ 0.5%	typical
Ollaya · SummEval relevance	47.5% ✎	▁··▁▂▁▅··▃██ beats 17/25	medium · avg 41%	▇ 0.5%	at its level
LeanMCP · helpsteer2 helpfulness	1.1446 ordinal levels ✎	▄▄▄▄▄·▄·█▄▄▄ beats 0/10	avg 1.357 ordinal levels	▇ 0.5%	blunt
DecisionPivot v1.1 · freshqa premise (auroc)	0.8320 auroc ✎	█·······█·██ beats 0/3	avg 0.209 auroc	▇ 0.4%	typical
DecideBench v1: returns policy	74.0% ✎	▃··▃··▃·▅▅█▃ beats 2/10	easy · avg 81%	▇ 0.4%	sharp · noisy
DecisionPivot v1.1 · contextual judge (accuracy)	57.5% ✎	█·██·······█ beats 0/3	medium · avg 43%	▇ 0.4%	at its level
LeanMCP · sst5	0.5321 ordinal levels ✎	▂·····▂▂▄█▄▂ beats 7/11	avg 0.879 ordinal levels	▇ 0.4%	blunt
Ollaya · PAWS paraphrase	81.2% ✎	▂▂▂▂·▃█▆·▂·▁ beats 13/25	easy · avg 81%	▇ 0.4%	typical
DecisionPivot v1.1 · cosapien policy (accuracy)	87.2% ✎	▄·······█··▄ beats 1/3	easy · avg 75%	▇ 0.4%	typical
DecisionPivot v1.1 · jbb judge (auroc)	0.9570 auroc ✎	█····█··█··█ beats 0/3	avg 0.299 auroc	▇ 0.4%	typical
DecisionPivot v1.1 · xstest (auroc)	0.9830 auroc ✎	█········▄·▄ beats 1/3	avg 0.320 auroc	▇ 0.4%	typical
Ollaya · Aegis 2 safety	78.8% ✎	▂·▁▁▂·▃▃▃█▃▁ beats 13/25	easy · avg 77%	▇ 0.4%	typical
LeanMCP · prompt injections	59.5% ✎	▂▃▂▂█·▄▂···▂ beats 1/11	medium · avg 62%	▇ 0.3%	at its level
DecisionPivot v1.1 · label errors (accuracy)	62.9% ✎	██·······█·█ beats 2/3	medium · avg 47%	▇ 0.3%	at its level
Ollaya · MASSIVE German intent	90.3% ✎	▁····▁·▂▃▆█▃ beats 22/23	easy · avg 84%	▇ 0.3%	sharp
JevalKit: JevBench public 231	82.7% ✎	█▄▄▄▄▄██▄··▄ beats 5/12	easy · avg 82%	▇ 0.3%	typical
DecisionPivot v1.1 · bfcl relevance (accuracy)	90.9% ✎	█··█······██ beats 2/3	easy · avg 82%	▇ 0.3%	typical
DecisionPivot v1.1 · jailbreak wild (accuracy)	77.3% ✎	█······█·█·█ beats 1/3	easy · avg 65%	▇ 0.3%	at its level
DecisionPivot v1.1 · ambigqa (accuracy)	61.3% ✎	█·······▄··▄ beats 2/3	medium · avg 44%	▇ 0.2%	at its level
Ollaya · SummEval consistency	81.9% ✎	▁·▁▁··▁▁▁▁▄█ beats 17/25	easy · avg 75%	▇ 0.2%	sharp · noisy
Ollaya · HelpSteer 2 scoring	38.1% ✎	▁▁··▁·▁▃▇█▅▄ beats 13/23	medium · avg 38%	▇ 0.2%	blunt · at its level
Ollaya · MASSIVE English intent	90.9% ✎	▁·······▂▃█▃ beats 22/23	very easy · avg 86%	▇ 0.2%	noisy
DecisionPivot v1.1 · selfaware (accuracy)	72.8% ✎	█····▄·····▄ beats 3/3	medium · avg 53%	▇ 0.1%	blunt · at its level
JevalKit: Typed Decisions held-out modal labels	70.2% ✎	▃▅··▃·▃▃·█▃█ beats 3/12	easy · avg 74%	▇ 0.0%	blunt
Ollaya · Civil Comments	90.0% ✎	▂▂▂·▂▅▅▆▃▅·█ beats 18/23	very easy · avg 86%	▇ 0.0%	blunt · noisy
LeanMCP · jailbreak classification	94.3% ✎	▂·▄··▃···█▆▂ beats 3/11	very easy · avg 95%	▇ 0.0%	blunt · maxed out · noisy

other tests5 not in a task · 2.8% of its score

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: private same skills (equated points)	43.4600 points ✎	▂▂▂▃▃▃▄▅▅▇█▃ beats 117/247	avg 41.206 points	▇ 1.2%	blunt
JevalKit: JevAlt-Bench EN/TR/DE	90.1% ✎	▅··▃▃▅▃█··▅▃ beats 7/12	very easy · avg 87%	▇ 0.7%	typical
Decision Index 0.4.1: private new domains (equated points)	45.8700 points ✎	▃█▃▃▂▆▃▃▄▄▄▄ beats 146/247	avg 38.861 points	▇ 0.6%	blunt
JevalKit: Decision Index public 0.3 (official weighted points)	55.4500 points ✎	▃▃·▃·▃▃▃▅▅·█ beats 7/12	avg 39.649 points	▇ 0.3%	blunt
JevalKit: Chinese decision validation v0.2	90.8% ✎	▃····█·█▃▃█▃ beats 10/12	very easy · avg 88%	▇ 0.0%	blunt

$ where to run it

Through an API

api · $0.038/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 ~24.6 GB, NVIDIA only · from huggingface.co/Cloudflare (official)
docker run --gpus all --shm-size 32g -p 30000:30000 -v ~/.cache/huggingface:/root/.cache/huggingface --ipc=host lmsysorg/sglang:dev-clef sglang serve --model-path Cloudflare/clef-flash --host 0.0.0.0 --port 30000

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 · DecisionPivot · StartLux Labs 32%, Decision Index 0.4.1 21%, LeanMCP · JevBench 14% and 9 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
DecisionPivot · StartLux Labs	48	▇▇▇▇▇▇▇▇▇▇ 32%	CC BY 4.0
Decision Index 0.4.1 · submitted; independent execution unverified	42	▇▇▇▇▇▇ 21%	CC BY 4.0
LeanMCP · JevBench	18	▇▇▇▇ 14%	MIT
System One Bench (Ali Alhashimi)	21	▇▇▇ 10%	CC-BY-4.0
Etsabary · decision benchmarks	2	▇▇ 7%	MIT
DecideBench (Cho Yin Yong)	8	▇ 4%	CC BY 4.0
Ollaya · original runs	14	▇ 4%	Apache-2.0
JevalKit (Valen Team)	8	▇ 4%	Apache-2.0
DecisionModelHub · support benchmarks	2	▇ 2%	CC-BY-4.0
Freyer et al. · MedQA	1	▇ 1%	CC BY 4.0
AgentDecisionBench	2	▇ 1%	CC-BY-4.0
Benchmark Heaven (own decision evaluations)	2	▇ 0%	CC BY 4.0
prior · pull toward the middle		▇ 0%
weighted average		= 57.6

not counted (3)

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]	34.4%	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
JevalKit (Valen Team)	JevalKit: VisualDecisionBench video [correct]	87.8%	fewer than three models have this test yet, or this source ran it for this model alone

$ 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