whatmodel.sh / decision / cua-s1-4b-0.2 (DecisionBench 16818868)

cua-s1-4b-0.2 (DecisionBench 16818868)

+ contribute

cua-s1-4b-0.2 (DecisionBench 16818868) is an open-weights decision model from cua-ai. Licence: apache-2.0.

Catalogue identity notes

DecisionBench v1 comparison: cua-ai/cua-s1-4b-0.2, revision 16818868b0cc7813808aae4e87b417657046ab79; language-model; adapter cua-s1-four-b-text-v1; probability source native_option_letter_logits_fp32_softmax. Kept separate from newer or unpinned Decision Index recipes; standard prompt, not compact. No unverified runtime or OpenRouter price imported. Current repository metadata checked separately; an unpinned alias is not evidence that today’s weights reproduce this historical run.

Model information sources

· Weights licence ↗

53.0±1.7#183 of 355 decision modelsfrom 27 benchmark results across 1 source

$ how it compares

The decision models ranked either side of cua-s1-4b-0.2 (DecisionBench 16818868), out of 357 with a score.

#	model	score	where	released
183	JevAny-Qwen-4B (direct token) (JevBench mini) SimpleJev	53±2	source code
184	decider-2b (DecisionBench fa996cea) Mapika	53±2	model files	2026-09-19
185	cua-s1-4b-0.2 (DecisionBench 16818868) cua-ai	53±2	model files
186	Bespoke-Nimble-9B (JevBench mini) Bespoke Labs	53±2	source code
187	Hopper 12B (trained) HopitAI	52±1	model files

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

$ what it's good at

workflows▁▁▁▂▁▂▂▆█▄▁▁ #62 of 114 · 99% of its score · 27 tests

Beats Clef-omni, Clef-flash, Bespoke Nimble 9B v3 here, though they rank higher overall. Trails InternLM Intern-Decision Qwen3.5-4B, StartLux-Decision v1.1 0.8B, JevK5 v0.3 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
DecisionBench v1 · action selection (all-request accuracy) submitted	54.9% ✎	▂▁▂▃▄▃▁▁▂█▇▄ beats 31/56	medium · avg 51%	▇▇▇▇▇▇▇▇ 20%	at its level
DecisionBench v1 · routing (all-request accuracy) submitted	68.0% ✎	▁▁▁▁▂▂▁▁▁▁▃█ beats 29/56	medium · avg 64%	▇▇▇▇▇ 14%	sharp · at its level
DecisionBench v1 · navigation (all-request accuracy) submitted	69.0% ✎	▁▁·▃▁▁▁▁▁▁▃█ beats 32/56	medium · avg 58%	▇▇▇▇ 11%	sharp · at its level
DecisionBench v1 · record classification (all-request accuracy) submitted	44.0% ✎	▃▂▁▂▂▂▄▁▃▆██ beats 37/56	medium · avg 39%	▇▇▇▇ 8.8%	at its level
DecisionBench v1 · verification (all-request accuracy) submitted	59.0% ✎	▁▁▅▁▃▂▂▁▁▇▇█ beats 33/56	medium · avg 56%	▇▇ 6.2%	blunt · at its level
DecisionBench v1 · document record classification (all-request accuracy) submitted	40.6% ✎	▂▂·▁▂▄▃▆▄█▁▁ beats 23/56	medium · avg 46%	▇▇ 5.0%	at its level
DecisionBench v1 · triage (all-request accuracy) submitted	84.0% ✎	▁·▁·▂▃▄▃▂█▄▂ beats 34/56	easy · avg 78%	▇▇ 4.6%	sharp
DecisionBench v1 · completion assessment (all-request accuracy) submitted	90.0% ✎	▁▁▂▃▄·▂▁▁▆█▂ beats 31/56	very easy · avg 86%	▇▇ 4.6%	sharp

19 more tests

test	result	vs the others in its run	difficulty	counts	why
DecisionBench v1 · rubric scoring prioritization (all-request accuracy) submitted	44.1% ✎	▁▁▅▅▃▃▅█▃▃·▁ beats 28/56	medium · avg 45%	▇▇ 4.2%	at its level
DecisionBench v1 · bounded extraction (all-request accuracy) submitted	92.2% ✎	▃▃▂·▁▁▁▁▁▁█▅ beats 27/56	very easy · avg 86%	▇▇ 4.1%	sharp · noisy
DecisionBench v1 · content classification (all-request accuracy) submitted	78.0% ✎	▁····▁▁▁▁▁▁█ beats 30/56	easy · avg 73%	▇ 3.4%	typical
DecisionBench v1 · review scoring (all-request accuracy) submitted	53.0% ✎	▁······▂▃▅▇█ beats 30/56	medium · avg 52%	▇ 2.8%	at its level
DecisionBench v1 · retrieval ranking (all-request accuracy) submitted	25.0% ✎	▁▁·▁▁▂▃▂▃▁▁█ beats 28/56	medium · avg 35%	▇ 2.2%	at its level · noisy
DecisionBench v1 · target selection (all-request accuracy) submitted	15.0% ✎	▁▁·▁▁▁▃▁▃▁▁█ beats 22/56	hard · avg 29%	▇ 2.0%	at its level · noisy
DecisionBench v1 · risk scoring (all-request accuracy) submitted	65.0% ✎	▁▁·▁▂▂▁▃▃█▇▃ beats 37/56	medium · avg 62%	▇ 1.9%	blunt · at its level
DecisionBench v1 · reasoning (all-request accuracy) submitted	47.7% ✎	▂·▁▁▆█▅▃▃▂▁▁ beats 33/56	medium · avg 52%	▇ 1.2%	at its level
DecisionBench v1 · retrieval verification (all-request accuracy) submitted	85.9% ✎	█▁▂▁▁▁▁▁▃▅▁▁ beats 39/56	easy · avg 74%	▇ 0.8%	noisy
DecisionBench v1 · document workflows (all-request accuracy) submitted	58.9% ✎	█▄▃▁▁▃▂▂▁·▁▁ beats 31/56	easy · avg 66%	▇ 0.7%	blunt · at its level
DecisionBench v1 · guardrails moderation (all-request accuracy) submitted	40.0% ✎	▁▁▁▂▆▃▃█···▁ beats 23/56	medium · avg 52%	▇ 0.5%	blunt · at its level · noisy
DecisionBench v1 · safety gating (all-request accuracy) submitted	91.0% ✎	▂·····▆█▂▃▄▇ beats 35/56	easy · avg 77%	▇ 0.4%	noisy
DecisionBench v1 · document classification (all-request accuracy) submitted	10.0% ✎	▂▁·▂▁▁▁▄▁▂▂█ beats 17/56	hard · avg 21%	▇ 0.3%	blunt · noisy
DecisionBench v1 · semantic filtering (all-request accuracy) submitted	83.0% ✎	▁·▁▁·▁▁█▂▃█▂ beats 42/56	easy · avg 71%	▇ 0.3%	blunt · at its level · noisy
DecisionBench v1 · content moderation (all-request accuracy) submitted	97.0% ✎	▁····▃▅▃▂▃·█ beats 37/56	easy · avg 82%	▇ 0.2%	noisy
DecisionBench v1 · context management (all-request accuracy) submitted	93.0% ✎	▁·····▁▇▆▁█▃ beats 51/56	easy · avg 73%	▇ 0.2%	blunt · at its level · noisy
DecisionBench v1 · entity alignment (all-request accuracy) submitted	50.1% ✎	▁·▁▄▇▄▅▄▂▁▄█ beats 22/56	easy · avg 66%	▇ 0.2%	at its level · noisy
DecisionBench v1 · function agent skill routing (all-request accuracy) submitted	49.0% ✎	▃▃▄▅▆▂▅▄▃▄█▃ beats 17/56	easy · avg 68%	▇ 0.1%	blunt · at its level · noisy
DecisionBench v1 · routing triage (all-request accuracy) submitted	0.0% ✎	█▁···▁·▂▆▅▂▂ beats 9/56	medium · avg 42%	▇ 0.0%	blunt · at its level · noisy

$ where to run it

Download model files ↗
Published model files verified. Includes adapters or readout weights; follow the model documentation for the required base model. This is the revision recorded by the benchmark. 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/cua-ai ↗

$ how the score was worked out · Hanno Labs DecisionBench v1 r… 99%

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
Hanno Labs DecisionBench v1 reviewed results · submitted; independent execution unverified	27	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 99%	CC0 1.0
prior · pull toward the middle		▇ 1%
weighted average		= 53.0

$ 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