whatmodel.sh / decision / reflex-27b (DecisionBench 1d4bf0f2)

reflex-27b (DecisionBench 1d4bf0f2)

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

reflex-27b (DecisionBench 1d4bf0f2) is an open-weights decision model from kshetrajna12. Its published size is 27.781427952B parameters. Licence: apache-2.0.

Catalogue identity notes

DecisionBench v1 comparison: kshetrajna12/reflex-27b, revision 1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0; language-model; adapter jev-compatible-server-v1; probability source two_order_option_letter_logit_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 ↗ · Parameter count ↗

Published served parameter count; not a RAM or download-size measurement.

62.7±1.8#80 of 355 decision modelsfrom 27 benchmark results across 1 source

$ how it compares

The decision models ranked either side of reflex-27b (DecisionBench 1d4bf0f2), out of 357 with a score, then more from kshetrajna12.

#modelscorewhereRAMreleased
78Nace.AI Drex v2.0 (35B-A3B) Nace.AI63±1model files
79Sieve-Vision-27B Sthānika AI63±4model files
80reflex-27b (DecisionBench 1d4bf0f2) kshetrajna1263±2model files
81decisio 0.6.0 (Gemma 4 31B) (decision readout) aminry63±1source code2026-10-05
82deck31b (Gemma 4 31B) (decision readout) krishna-gogineni-76563±1source code
more from kshetrajna12
105reflex-qwen3.5-4b-lora (DecisionBench 9df1cbbb) kshetrajna1261±2model files
106reflex Qwen3.8-27B-FP8 (pmscr) (decision readout) kshetrajna1261±1source code

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

$ what it's good at

workflows▁▁▁▂▁▂▂▆█▄▁▁ #13 of 114 · 99% of its score · 27 tests

Beats Nace.AI Drex Max v1 (27B), Perplexity Decider v1.2 (27B), d1-ultra here, though they rank higher overall. Trails GPT-6 Luna Decisions here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
DecisionBench v1 · action selection (all-request accuracy) submitted65.4% ✎▂▁▂▃▄▃▁▁▂█▇▄ beats 53/56medium · avg 51%▇▇▇▇▇▇▇▇▇ 21%at its level
DecisionBench v1 · routing (all-request accuracy) submitted78.0% ✎▁▁▁▁▂▂▁▁▁▁▃█ beats 55/56medium · avg 64%▇▇▇▇▇ 12%sharp
DecisionBench v1 · record classification (all-request accuracy) submitted52.0% ✎▃▂▁▂▂▂▄▁▃▆██ beats 56/56medium · avg 39%▇▇▇▇ 11%at its level
DecisionBench v1 · navigation (all-request accuracy) submitted72.0% ✎▁▁·▃▁▁▁▁▁▁▃█ beats 45/56medium · avg 58%▇▇▇▇ 10%sharp
DecisionBench v1 · verification (all-request accuracy) submitted64.5% ✎▁▁▅▁▃▂▂▁▁▇▇█ beats 55/56medium · avg 56%▇▇▇ 6.9%blunt · at its level
DecisionBench v1 · document record classification (all-request accuracy) submitted56.3% ✎▂▂·▁▂▄▃▆▄█▁▁ beats 50/56medium · avg 46%▇▇ 5.7%at its level
DecisionBench v1 · rubric scoring prioritization (all-request accuracy) submitted55.4% ✎▁▁▅▅▃▃▅█▃▃·▁ beats 47/56medium · avg 45%▇▇ 4.9%at its level
DecisionBench v1 · triage (all-request accuracy) submitted86.0% ✎▁·▁·▂▃▄▃▂█▄▂ beats 41/56easy · avg 78%▇ 3.5%sharp
19 more tests
testresultvs the others in its rundifficultycountswhy
DecisionBench v1 · review scoring (all-request accuracy) submitted61.0% ✎▁······▂▃▅▇█ beats 45/56medium · avg 52%▇ 3.1%at its level
DecisionBench v1 · completion assessment (all-request accuracy) submitted95.0% ✎▁▁▂▃▄·▂▁▁▆█▃ beats 53/56very easy · avg 86%▇ 2.9%sharp · maxed out
DecisionBench v1 · content classification (all-request accuracy) submitted82.0% ✎▁····▁▁▁▁▁▁█ beats 55/56easy · avg 73%▇ 2.9%typical
DecisionBench v1 · retrieval ranking (all-request accuracy) submitted61.0% ✎▁▁·▁▁▂▃▂▃▁▁█ beats 54/56medium · avg 35%▇ 2.6%at its level · noisy
DecisionBench v1 · target selection (all-request accuracy) submitted48.0% ✎▁▁·▁▁▁▃▁▂▁▁█ beats 49/56hard · avg 29%▇ 2.6%at its level · noisy
DecisionBench v1 · bounded extraction (all-request accuracy) submitted94.9% ✎▃▃▂·▁▁▁▁▁▁█▅ beats 46/56very easy · avg 86%▇ 2.1%sharp · maxed out · noisy
DecisionBench v1 · risk scoring (all-request accuracy) submitted64.0% ✎▁▁·▁▂▂▁▃▃█▇▃ beats 30/56medium · avg 62%▇ 2.0%blunt · at its level
DecisionBench v1 · reasoning (all-request accuracy) submitted74.8% ✎▂·▁▁▆█▅▃▃▃▁▁ beats 52/56medium · avg 52%▇ 1.3%at its level
DecisionBench v1 · document workflows (all-request accuracy) submitted82.7% ✎█▄▃▁▁▃▂▃▁·▁▁ beats 51/56easy · avg 66%▇ 0.8%blunt · at its level
DecisionBench v1 · retrieval verification (all-request accuracy) submitted88.2% ✎█▁▂▁▁▁▁▁▂▅▁▁ beats 51/56easy · avg 74%▇ 0.7%noisy
DecisionBench v1 · guardrails moderation (all-request accuracy) submitted62.9% ✎▁▁▁▂▆▃▃█···▁ beats 44/56medium · avg 52%▇ 0.6%blunt · at its level · noisy
DecisionBench v1 · document classification (all-request accuracy) submitted3.0% ✎▂▁·▂▃▁▁▄▁▂▂█ beats 9/56hard · avg 21%▇ 0.4%blunt · at its level · noisy
DecisionBench v1 · safety gating (all-request accuracy) submitted98.0% ✎▂·····▆█▂▂▄▇ beats 54/56easy · avg 77%▇ 0.4%noisy
DecisionBench v1 · semantic filtering (all-request accuracy) submitted86.0% ✎▁·▁▁·▁▁█▂▃█▂ beats 48/56easy · avg 71%▇ 0.3%blunt · noisy
DecisionBench v1 · context management (all-request accuracy) submitted89.5% ✎▁·····▁▇▆▁█▃ beats 42/56easy · avg 73%▇ 0.2%blunt · noisy
DecisionBench v1 · content moderation (all-request accuracy) submitted100.0% ✎▁····▃▅▃▂▂·█ beats 50/56easy · avg 82%▇ 0.2%noisy
DecisionBench v1 · entity alignment (all-request accuracy) submitted27.4% ✎▁·▁▄▇▄▅▄▂▁▄█ beats 14/56easy · avg 66%▇ 0.2%noisy
DecisionBench v1 · function agent skill routing (all-request accuracy) submitted27.1% ✎▃▃▄▅▆▂▅▄▃▄█▃ beats 11/56easy · avg 68%▇ 0.1%blunt · noisy
DecisionBench v1 · routing triage (all-request accuracy) submitted0.0% ✎█▁···▁·▂▆▅▂▂ beats 3/56medium · avg 42%▇ 0.0%blunt · at its level · noisy

$ where to run it

Download model files ↗
Published model files verified. 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/Qwen ↗

$ 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

sourceresultsshare of the scorelicence
Hanno Labs DecisionBench v1 reviewed results · submitted; independent execution unverified27CC0 1.0
prior · pull toward the middle
weighted average= 62.7

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit 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 · sources and licences · json