whatmodel.sh / decision / CLM-v0.1-8B Qwen3-8B heads

CLM-v0.1-8B Qwen3-8B heads

+ contribute

CLM-v0.1-8B Qwen3-8B heads is an open-weights decision model from Contrastive-LM. Its published size is 8.19B parameters. Licence: apache-2.0.

Catalogue identity notes

Exact Decision Index tested engine clm-v0.1-8b; reported model CLM-v0.1-8B Qwen3-8B heads. Type: full fine-tune; technique: autoregressive. Published checkpoint/base reference: Contrastive-LM/CLM-v0.1-8B. Implementation: https://github.com/Contrastive-LM/CLM. 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: clm:8b, manifest 3f89a66972f1398210d2f41b94667665132aa51c2d313237c3fc70a49c275cb5, matching the independent accuracy-run manifest. Native family implementation; published runtime precision label F32. Memory is measured peak GPU increase (15678 MiB, rounded upward to 15.4 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.

20.4±0.8#341 of 355 decision modelsfrom 72 benchmark results across 5 sources

$ how it compares

The decision models ranked either side of CLM-v0.1-8B Qwen3-8B heads, out of 357 with a score, then more from Contrastive-LM.

#	model	score	where	RAM
341	Verdict heman10x	21±1	model files
342	jeff GLiFormer-large (decision readout) logan-markewich	21±1	source code
343	CLM-v0.1-8B Qwen3-8B heads Contrastive-LM	20±1	local	~15GB
344	gliner2.5-base-v1 (DecisionBench 1a8bc24e) Fastino	20±2	model files
345	gliner2.5-multi-v1 (DecisionBench a221b77a) Fastino	20±2	model files
	more from Contrastive-LM
290	CLM v0.1 (JevBench mini) Contrastive-LM	33±2	source code

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

$ what it's good at

tools & automation▁▃▁▁▁▂▃▄▆██▆ #237 of 250 · 16% of its score · 10 tests

Beats DM-JEPA 1.1, LFM2.5-2.6B-RLCD, Von 1.3 here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
DecideBench v1: agent routing	34.0% ✎	▄·····▂▄·█·▄ beats 0/10	very easy · avg 93%	▇▇ 5.2%	sharp · at its level · noisy
Decision Index 0.4.1: BFCL (case exact accuracy) submitted	38.9% ✎	▁▃▁▁▁▂▂▄▅▆█▁ beats 15/246	very easy · avg 92%	▇▇ 4.3%	sharp · at its level · noisy
DecideBench v1: action review	50.0% ✎	▃▃·▃·█···█·▅ beats 1/10	easy · avg 87%	▇ 2.0%	sharp · at its level · noisy
AgentDecisionBench: L1 BFCL v4 + When2Call (3,991 items)	46.2% ✎	█·····▄·▄▄·▄ beats 0/5	easy · avg 75%	▇ 1.7%	at its level
Decision Index 0.4.1: API-Bank (accuracy) submitted	6.1% ✎	▁▁▁▃▁▁▁▂▃▂█▂ beats 13/236	easy · avg 70%	▇ 1.0%	sharp · noisy
Decision Index 0.4.1: When2Call (accuracy) submitted	38.7% ✎	▁▁▁▃▃▅▅▆█▇█▃ beats 25/246	easy · avg 72%	▇ 0.6%	at its level · noisy
AgentDecisionBench: L2 DecisionGym common states (200 tasks)	11.9% ✎	██···█·██··█ beats 0/5	medium · avg 57%	▇ 0.5%	sharp · noisy
Decision Index 0.4.1: ToolRet (nDCG@10) submitted	40.3% ✎	▁▁▁▁▁·▃▁▁▄█▁ beats 20/246	medium · avg 60%	▇ 0.5%	blunt · at its level

2 more tests

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: Home appliance simulator (case exact accuracy) submitted	0.0% ✎	█▁▁▃▄▄▆▄▁▂▁▃ beats 25/243	medium · avg 37%	▇ 0.0%	sharp · too hard for it · noisy
Decision Index 0.4.1: RouterBench (selected quality (quality objective)) submitted	39.3% ✎	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▃▁▁·▁▁▁▁▃▃█▅ #258 of 258 · 5.1% of its score · 8 tests

Trails Zev (decision readout), LFM2.5-350M-RLCD, Nirnaya 0.4B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
DecideBench v1: support intent	46.0% ✎	▃····▂▂·▃··█ beats 0/10	very easy · avg 95%	▇ 2.3%	sharp · at its level · noisy
Decision Index 0.4.1: Amazon ESCI (macro-F1) submitted	5.0% ✎	▃▁▁▁▁▁▂▂▃▃█▅ beats 0/246	medium · avg 47%	▇ 0.8%	at its level
Decision Index 0.4.1: BRIGHT (nDCG@10) submitted	29.1% ✎	▁▁▁▁▁▁▁▃▂▄▇█ beats 44/246	medium · avg 38%	▇ 0.7%	blunt
Decision Index 0.4.1: CLINC150+OOS (macro-F1) submitted	8.9% ✎	▁▁▁▃▁▁▁▂▃▆█▄ beats 9/238	easy · avg 83%	▇ 0.4%	sharp · at its level · noisy
Decision Index 0.4.1: HoVer (accuracy) submitted	52.0% ✎	▆▆▇▅▄▄▃▇▄█▆▄ beats 15/246	easy · avg 74%	▇ 0.4%	blunt · at its level
Decision Index 0.4.1: PhishNChips phishing decisions (accuracy) submitted	65.0% ✎	█▅▅▄▃▂▁▁▁▁·▁ beats 105/245	easy · avg 70%	▇ 0.3%	at its level · noisy
Decision Index 0.4.1: BANKING77 (macro-F1) submitted	6.5% ✎	▁·▁▃▁▁▁▁▂█▇█ beats 4/238	easy · avg 79%	▇ 0.2%	at its level · noisy
Decision Index 0.4.1: SGD/SGD-X (macro-F1) submitted	10.0% ✎	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▂▃▂▃▂▅▆▃█▆▄▃ #236 of 256 · 20% of its score · 12 tests

Beats Crivo-68M, Bosun v3.1 1.7B LoRA, Mola-68M here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: HellaSwag (accuracy) submitted	40.4% ✎	▂▄▂▂▄▇▃▇█▅▃▁ beats 20/246	very easy · avg 85%	▇▇ 5.6%	sharp · at its level
DecideBench v1: review sentiment	44.0% ✎	▂··▃···▂▂█▄▂ beats 1/10	easy · avg 85%	▇ 3.7%	sharp
Decision Index 0.4.1: NLI4CT (macro-F1) submitted	56.5% ✎	▁▁▁▂▃▃▃▄▇▇█▄ beats 29/246	easy · avg 74%	▇ 3.5%	at its level
Decision Index 0.4.1: ANLI (macro-F1) submitted	31.6% ✎	▁▃▃▅▅█▂····▁ beats 19/246	medium · avg 60%	▇ 2.1%	at its level
DecideBench v1: content moderation	22.0% ✎	▃··▂···▄▄▂·█ beats 0/10	easy · avg 83%	▇ 1.9%	sharp · noisy
DecideBench v1: claim support	44.0% ✎	▄··▄·▄·▄▄███ beats 1/10	easy · avg 82%	▇ 0.9%	sharp · noisy
Decision Index 0.4.1: FinEntity (macro-F1) submitted	33.4% ✎	▁···▁▁▃▁▃▆█▃ beats 9/246	easy · avg 84%	▇ 0.7%	at its level · noisy
Decision Index 0.4.1: ContractNLI (macro-F1) submitted	20.1% ✎	▁▁▃▁▁▁▁▂▂▇▇█ beats 4/246	easy · avg 71%	▇ 0.6%	at its level · noisy

4 more tests

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: iSarcasmEval (Sarcasm F1 · track A, English) submitted	31.2% ✎	▁·▁·▁▁▂▄▅█▇▄ beats 64/246	medium · avg 44%	▇ 0.6%	typical
Decision Index 0.4.1: VAST (macro-F1) submitted	22.7% ✎	▁▃▁▂▅▄▄▂▄▆█▄ beats 2/246	medium · avg 62%	▇ 0.5%	at its level
Decision Index 0.4.1: RAGTruth (F1 on hallucinated class) submitted	51.8% ✎	▁▁▁▁▁▁▁▁▁▃▆█ beats 51/246	easy · avg 66%	▇ 0.4%	at its level · noisy
Decision Index 0.4.1: ACOS (per-review F1) submitted	3.1% ✎	▁··▃▃▃▅▇█▇▃▃ beats 8/246	hard · avg 22%	▇ 0.1%	too hard for it · noisy

arts & human taste▃▇▄▃▅▅█▅▄▂▁▁ #247 of 247 · 4.4% of its score · 6 tests

Trails Zev (decision readout), LFM2.5-350M-RLCD, Nirnaya 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	25.4% ✎	▁▃▂▃▃▃▇▅█▅▃▁ beats 11/246	medium · avg 61%	▇ 1.6%	at its level
Decision Index 0.4.1: BPoMP (accuracy) submitted	32.9% ✎	▃█▄▄▅█▆▅▄▁▁▁ beats 0/246	easy · avg 83%	▇ 1.2%	sharp · at its level · noisy
Decision Index 0.4.1: Humicroedit (accuracy) submitted	55.6% ✎	▁·▁▃▂▃▅█▄▂▁▁ beats 62/246	medium · avg 60%	▇ 0.9%	blunt · at its level
Decision Index 0.4.1: cfcolor (accuracy) submitted	50.8% ✎	▁········▁▇█ beats 24/246	medium · avg 60%	▇ 0.5%	blunt · at its level
Decision Index 0.4.1: Habermas Machine (accuracy) submitted	37.1% ✎	▁▂▄█▇▅▁▁▄▄▁▁ beats 31/246	medium · avg 47%	▇ 0.1%	blunt · at its level
Decision Index 0.4.1: POP909-CL (accuracy) submitted	0.0% ✎	▄▅▅▃▆▅▃▅▂▅▃█ beats 9/237	hard · avg 21%	▇ 0.0%	sharp · too hard for it · noisy

knowledge & reasoning▁▃▂▂▂▃▃▅▅▆█▂ #245 of 263 · 34% of its score · 19 tests

Beats Amber 0.4B, Dinah-1, Crivo-68M here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: ARC-Easy (accuracy) submitted	56.4% ✎	▁▁▃▁▂▂▂▂▃▆█▅ beats 16/246	very easy · avg 96%	▇▇▇▇ 10%	sharp · at its level · noisy
Decision Index 0.4.1: ARC-Challenge (accuracy) submitted	41.5% ✎	▂▃▂▂▂▂▁▃▄▆█▆ beats 19/246	very easy · avg 90%	▇▇ 6.2%	sharp · at its level
Decision Index 0.4.1: MMLU (accuracy) submitted	35.8% ✎	▁▄▄▃▃▂▅▇▅█▇▁ beats 23/246	easy · avg 70%	▇ 3.6%	sharp · at its level
Decision Index 0.4.1: BBH (accuracy) submitted	30.8% ✎	▄▆▄▄█▇▇▄▃▁▁▁ beats 2/246	easy · avg 66%	▇ 2.9%	at its level
Decision Index 0.4.1: WinoGrande (accuracy) submitted	50.8% ✎	█▅█▅▄█▆▄▃▃▁▁ beats 14/246	easy · avg 78%	▇ 2.4%	at its level
Decision Index 0.4.1: MMLU-Pro (accuracy) submitted	14.7% ✎	▄▄▃▃▂▄▅▇██▅▁ beats 18/246	medium · avg 46%	▇ 2.1%	sharp
Ollaya · BoolQ	56.3% ✎	▃··▃▃▁▄▁▃█▇▄ beats 0/25	easy · avg 84%	▇ 1.4%	at its level
Ollaya · PubMedQA	40.0% ✎	▃▂▃▃·▂▃▃▆█▆▂ beats 1/25	easy · avg 69%	▇ 1.2%	at its level

11 more tests

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: GPQA Diamond (accuracy) submitted	20.4% ✎	▃▆▅▅▆█▆▃▁··▁ beats 1/246	medium · avg 39%	▇ 1.0%	blunt
Ollaya · SQuAD 2 answerability	52.2% ✎	▃▁▃▁▇█▃▃▄▃·▁ beats 0/25	easy · avg 80%	▇ 0.6%	blunt · at its level
Decision Index 0.4.1: GSM8K (accuracy (4- and 10-choice)) submitted	17.5% ✎	▁··▄█▆▆▆▅▅▃▁ beats 7/246	medium · avg 45%	▇ 0.6%	sharp
Decision Index 0.4.1: MuSR (accuracy) submitted	34.6% ✎	▃▁▂█▅▄▁▁▁▁▁▁ beats 0/246	medium · avg 60%	▇ 0.6%	blunt · at its level
Decision Index 0.4.1: CLadder (accuracy) submitted	53.9% ✎	▁███▆▅▃▃▁▁▁▁ beats 32/246	easy · avg 71%	▇ 0.6%	at its level
Ollaya · VitaminC verification	38.7% ✎	▃···▃·▄█▃··▁ beats 0/25	easy · avg 78%	▇ 0.6%	blunt · at its level
Decision Index 0.4.1: CRUXEval (accuracy) submitted	42.8% ✎	██▄▃▃▄▁▁···▁ beats 74/246	medium · avg 58%	▇ 0.4%	at its level · noisy
Ollaya · MultiNLI	66.6% ✎	▃··▄██▂▆▆█▆▂ beats 0/25	very easy · avg 87%	▇ 0.4%	blunt
Decision Index 0.4.1: SATA-Bench (case exact accuracy) submitted	5.6% ✎	▁▁▁▁▁▃▁▂▅█▂▁ beats 35/246	hard · avg 22%	▇ 0.2%	too hard for it
Decision Index 0.4.1: ChessBench (accuracy) submitted	8.2% ✎	▄▇▄▄▄▆▄█▄▁▁▁ beats 26/246	hard · avg 16%	▇ 0.2%	blunt
Decision Index 0.4.1: HLE (accuracy) submitted	15.8% ✎	▁▁▂▃▆█▇▅▅▃▁▁ beats 215/246	very hard · avg 13%	▇ 0.0%	blunt

workflows▁▃▁▂▁▂▂▆█▄▁▁ #111 of 114 · 16% of its score · 18 tests

Trails KaLM-Reranker-V1-Nano-R2 (DecisionBench 3902d645), KaLM-Reranker-V1-Small-R2 (DecisionBench d1c8fdf9), bge-reranker-v2-m3 (DecisionBench 953dc6f6) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
DecideBench v1: ticket triage	42.0% ✎	▄·····▂█·▆·▂ beats 1/10	very easy · avg 90%	▇▇ 5.6%	sharp · at its level
Ollaya · Typed decisions test	35.7% ✎	▄·▂▆▄▄▂·▄█▆▆ beats 0/22	medium · avg 64%	▇ 1.7%	at its level
Ollaya · PAWS paraphrase	51.6% ✎	▃▂▂▂·▃█▆·▂·▁ beats 1/25	easy · avg 81%	▇ 1.6%	at its level
Ollaya · MASSIVE German intent	6.0% ✎	▃····▁·▂▃▆█▃ beats 0/23	easy · avg 84%	▇ 1.6%	sharp · at its level
DecideBench v1: returns policy	46.0% ✎	▃··▃··▃·▅▅█▃ beats 1/10	easy · avg 81%	▇ 1.4%	sharp · at its level · noisy
Ollaya · Aegis 2 safety	42.4% ✎	▃·▁▁▂·▃▃▂█▃▁ beats 0/25	easy · avg 77%	▇ 1.2%	at its level
Ollaya · MASSIVE English intent	8.0% ✎	▃·······▂▃█▃ beats 0/23	very easy · avg 86%	▇ 0.9%	at its level · noisy
Ollaya · SummEval consistency	1.4% ✎	▃·▁▁··▁▁▁▁▄█ beats 0/25	easy · avg 75%	▇ 0.5%	sharp · noisy

10 more tests

test	result	vs the others in its run	difficulty	counts	why
Ollaya · SummEval relevance	22.9% ✎	▁··▁▂▁▅··▃██ beats 7/25	medium · avg 41%	▇ 0.5%	typical
Ollaya · HelpSteer 2 scoring	19.3% ✎	▁▃··▁·▁▃▇█▅▄ beats 1/23	medium · avg 38%	▇ 0.3%	blunt · at its level
Perch semantic lint · private (Youden J)	0.0141 J ✎	█····▄█▄···█ beats 1/7	avg 0.380 J	▇ 0.2%	blunt
Perch semantic lint · public (Youden J)	0.0140 J ✎	█····▄█▄···█ beats 1/7	avg 0.363 J	▇ 0.2%	blunt
Ollaya · Civil Comments	84.3% ✎	▂▂▂·▂▅▅▆▃▅·█ beats 8/23	very easy · avg 86%	▇ 0.1%	blunt · noisy
Perch security detection · public (Youden J)	0.0000 J ✎	█···▄▄···▄▄█ beats 0/7	avg 0.052 J	▇ 0.0%	blunt
Perch bug detection · private (Youden J)	-0.0059 J ✎	▅··▃▃█·····▃ beats 0/7	avg 0.027 J	▇ 0.0%	blunt
Perch bug detection · public (Youden J)	0.0102 J ✎	█▄·█·▄▄····▄ beats 1/7	avg 0.026 J	▇ 0.0%	blunt
Perch security detection · private (Youden J)	0.0000 J ✎	█▃▃··▃·▃···▃ beats 1/7	avg 0.017 J	▇ 0.0%	blunt
Structured state reasoning · V3 probe	38.3% ✎	█··········█ beats 0/1		—	not counted yet: no other model in the same run to compare it with

other tests2 not in a task · 3.7% of its score

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: private same skills (equated points)	4.9800 points ✎	▃▂▂▃▃▃▄▅▅▇█▃ beats 8/247	avg 41.206 points	▇ 2.4%	blunt
Decision Index 0.4.1: private new domains (equated points)	8.7500 points ✎	▃█▃▃▂▆▃▃▄▄▄▄ beats 13/247	avg 38.861 points	▇ 1.3%	blunt

$ where to run it

Other ways to run it

Run it as published · needs 15.4 GB, NVIDIA only · from github.com/ollaya-dev
ollaya run clm:8b --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/Contrastive-LM ↗

$ how the score was worked out · Decision Index 0.4.1 61%, DecideBench (Cho Yin Yong) 23%, Ollaya · original runs 13% and 2 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	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 61%	CC BY 4.0
DecideBench (Cho Yin Yong)	8	▇▇▇▇▇▇▇ 23%	CC BY 4.0
Ollaya · original runs	14	▇▇▇▇ 13%	Apache-2.0
AgentDecisionBench	2	▇ 2%	CC-BY-4.0
Perch benchmark results	6	▇ 1%	CC BY 4.0
prior · pull toward the middle		▇ 0%
weighted average		= 20.4

not counted (3)

Decision Index 0.4.1	Decision Index 0.4.1: RouterBench (selected quality (quality objective)) [correct]	39.3%	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]	10.0%	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
Structured state reasoning probe	Structured state reasoning · V3 probe [correct]	38.3%	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