JevBench / index.html
florian-chutes's picture
Clarify JevBench decision model identity, method and snapshot dates
0f462da verified
Raw History Blame Contribute Delete
44.5 kB
<!doctype html>
<html lang="en" data-theme="dark">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width,initial-scale=1">
<title>JevBench — AI Decision Model Benchmark | Benchmark Heaven</title>
<meta name="description" content="JevBench by Benchmark Heaven evaluates AI decision models and Jev-compatible systems on typed accuracy, calibration, latency and cost. Static snapshot captured 2026-10-05; current board at benchmarkheaven.com/jev-models.">
<link rel="canonical" href="https://benchmarkheaven.com/jev-models">
<meta property="og:type" content="website">
<meta property="og:url" content="https://huggingface.co/spaces/benchmarkheaven/JevBench">
<meta property="og:site_name" content="JevBench by Benchmark Heaven">
<meta property="og:title" content="JevBench — AI decision model benchmark">
<meta property="og:description" content="JevBench by Benchmark Heaven evaluates AI decision models and Jev-compatible systems on typed accuracy, calibration, latency and cost. Static snapshot captured 2026-10-05; current board at benchmarkheaven.com/jev-models.">
<meta name="twitter:card" content="summary">
<meta name="twitter:title" content="JevBench — AI decision model benchmark">
<meta name="twitter:description" content="JevBench by Benchmark Heaven evaluates AI decision models and Jev-compatible systems on typed accuracy, calibration, latency and cost. Static snapshot captured 2026-10-05; current board at benchmarkheaven.com/jev-models.">
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "Dataset",
"name": "JevBench — AI decision model benchmark",
"description": "JevBench by Benchmark Heaven is a benchmark for AI decision models, including Jev-compatible systems. It measures typed decision accuracy, probability calibration, latency and cost. Maintained by Benchmark Heaven, independently of TypeSafe AI. Capability is the mean of Intelligence and Calibration, ranked within official cost and median-latency caps. Composite: Weighted harmonic mean of Intelligence, Calibration, Speed and Cost, multiplied by squared penalties when Intelligence is below its configured floor, or Speed or Cost is below 50. Equal 25% weights do not mean an arithmetic average.",
"keywords": "AI decision model benchmark, decision models, leaderboard, model evaluation, calibration, JevBench, Jev-compatible, open weights",
"license": "https://creativecommons.org/licenses/by/4.0/",
"url": "https://benchmarkheaven.com/jev-models",
"sameAs": [
"https://huggingface.co/spaces/benchmarkheaven/JevBench",
"https://huggingface.co/datasets/benchmarkheaven/jevbench"
],
"creator": {
"@type": "Organization",
"name": "Benchmark Heaven",
"url": "https://benchmarkheaven.com"
},
"variableMeasured": [
"Intelligence",
"Calibration",
"Speed",
"Cost",
"JevBench Capability Score",
"Composite score"
],
"distribution": [
{
"@type": "DataDownload",
"name": "leaderboard.csv",
"url": "https://huggingface.co/datasets/benchmarkheaven/jevbench",
"encodingFormat": "text/csv"
}
],
"dateModified": "2026-10-05"
}
</script>
<link rel="stylesheet" href="style.css">
</head>
<body>
<main>
<header class="top">
<span class="wordmark">Benchmark Heaven</span>
<button class="theme-toggle" id="themeToggle" type="button">Toggle light / dark</button>
</header>
<div class="tag">Open benchmark</div>
<h1>JevBench — AI Decision Model Benchmark</h1>
<p class="lede">JevBench by Benchmark Heaven is a benchmark for AI decision models, including Jev-compatible systems. It measures typed decision accuracy, probability calibration, latency and cost. Maintained by Benchmark Heaven, independently of TypeSafe AI. This fully static, crawlable snapshot preserves the v1.6.1 release feed captured on 2026-10-05; the table below is rendered at build time, not loaded by JavaScript. Capability is the mean of Intelligence and Calibration, ranked within the official cost and median-latency caps. Composite is a weighted harmonic mean with configured penalties; its equal 25% weights do not make it an arithmetic average.</p>
<div class="actions">
<a class="btn primary" href="https://benchmarkheaven.com/jev-models" target="_blank" rel="noopener">Live interactive board</a>
<a class="btn" href="https://github.com/fstandhartinger/jevbench" target="_blank" rel="noopener">Method &amp; code</a>
<a class="btn" href="https://benchmarkheaven.com/submit" target="_blank" rel="noopener">Submit a system</a>
</div>
<div class="facts">
<div class="fact"><b>92</b><span>ranked systems</span></div>
<div class="fact"><b>94</b><span>systems listed in total</span></div>
<div class="fact"><b>4</b><span>axes · equal 25% weights in Composite</span></div>
<div class="fact"><b>v1.6.1</b><span>snapshot captured 2026-10-05</span></div>
</div>
<h2>Current leaderboard — JevBench Capability Score</h2>
<p class="note">Data: JevBench v1.6.1, generated 2026-10-05 · 92 ranked systems · only system-level aggregates are published. Rows inside the official caps first, in Capability rank order; the other ranked rows follow, in Composite order. Click a column header to sort.</p>
<div class="table-wrap">
<table id="board">
<thead>
<tr>
<th data-type="num">Capability rank</th>
<th>Model</th>
<th data-type="num">Capability score</th>
<th data-type="num">Intelligence</th>
<th data-type="num">Calibration</th>
<th data-type="num">Speed</th>
<th data-type="num">Cost axis</th>
<th data-type="num">USD per 1,000 decisions</th>
<th data-type="num">p50 latency s (raw)</th>
</tr>
</thead>
<tbody>
<tr><span class="rank">1.</span><td><a class="sys-name" href="https://huggingface.co/chinhnc/Quyet-1.0-Large" target="_blank" rel="noopener">Quyet-1.0-Large (Chinh Nguyen, Gemma-4-31B decoder, option-letter logits)</a></td><td class="score">81.7</td><td>73.4</td><td>90.0</td><td>86.9</td><td>50.5</td><td>$0.0445 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">2.</span><td><a class="sys-name" href="https://levanto.ai" target="_blank" rel="noopener">Sage 1.3.0 (Levanto Labs)</a></td><td class="score">78.6</td><td>65.6</td><td>91.6</td><td>93.6</td><td>58.2</td><td>$0.0247 <span class="est">est.</span></td><td>0.2</td></tr>
<tr><span class="rank">3.</span><td><a class="sys-name" href="https://github.com/krishna-gogineni-765/deck31b" target="_blank" rel="noopener">deck-31B (krishna765, frozen Gemma-4-31B-it, TorchAO FP8 dynamic)</a></td><td class="score">77.6</td><td>73.0</td><td>82.2</td><td>86.7</td><td>49.7</td><td>$0.0475 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">4.</span><td><a class="sys-name" href="https://docs.typesafe.ai" target="_blank" rel="noopener">Jev 1.13.0 (TypeSafe AI)</a></td><td class="score">77.1</td><td>63.6</td><td>90.6</td><td>91.5</td><td>54.7</td><td>$0.0323</td><td>0.2</td></tr>
<tr><span class="rank">5.</span><td><a class="sys-name" href="https://huggingface.co/torchcast-ai/torchcast-decision-12b" target="_blank" rel="noopener">torchcast-decision-12b (Torchcast AI, Gemma-4-12B fine-tune, option-letter logprob readout)</a></td><td class="score">71.7</td><td>60.5</td><td>82.9</td><td>91.7</td><td>56.4</td><td>$0.0285 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">6.</span><td><a class="sys-name" href="https://huggingface.co/akhilaaa3/Jev-Omni" target="_blank" rel="noopener">Jev-Omni (akhilaaa3, Gemma-4-12B merged)</a></td><td class="score">71.3</td><td>55.5</td><td>87.0</td><td>85.4</td><td>56.1</td><td>$0.0292 <span class="est">est.</span></td><td>0.2</td></tr>
<tr><span class="rank">7.</span><td><a class="sys-name" href="https://huggingface.co/EldanRing/Winnow-12B" target="_blank" rel="noopener">Winnow-12B Q8</a></td><td class="score">71.2</td><td>59.5</td><td>83.0</td><td>86.7</td><td>56.6</td><td>$0.0281 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">8.</span><td><a class="sys-name" href="https://github.com/blockbrain-ai/cygnet-recipe" target="_blank" rel="noopener">Cygnet (blockbrain, frozen Gemma-4-12B-it)</a></td><td class="score">70.9</td><td>54.8</td><td>87.0</td><td>91.8</td><td>56.4</td><td>$0.0283 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">9.</span><td><a class="sys-name" href="https://huggingface.co/crh225/plumb-4b" target="_blank" rel="noopener">Plumb-4B (crh225, JevK5 v0.2 + LoRA)</a></td><td class="score">65.4</td><td>43.0</td><td>87.9</td><td>93.8</td><td>63.1</td><td>$0.017 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">10.</span><td><a class="sys-name" href="https://github.com/Mapika/decider" target="_blank" rel="noopener">decider-4b v2 (Mapika)</a></td><td class="score">64.9</td><td>40.1</td><td>89.6</td><td>91.8</td><td>64.5</td><td>$0.0152 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">11.</span><td><a class="sys-name" href="https://huggingface.co/alibiserikbay/JevK5" target="_blank" rel="noopener">JevK5 v0.3 (4B)</a></td><td class="score">64.2</td><td>38.5</td><td>89.9</td><td>93.9</td><td>63.1</td><td>$0.017 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">12.</span><td><a class="sys-name" href="https://huggingface.co/Cloudflare/clef-flash" target="_blank" rel="noopener">Clef-Flash (Cloudflare, Qwen3.5-9B post-train with a joint schema head, multimodal, measured on text)</a></td><td class="score">63.8</td><td>42.2</td><td>85.5</td><td>88.4</td><td>46.8</td><td>$0.0593 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">13.</span><td><a class="sys-name" href="https://github.com/krishna-gogineni-765/deck" target="_blank" rel="noopener">deck-4B v1.0 (krishna765, JevK5 + LoRA, FP8 weight-only)</a></td><td class="score">63.6</td><td>38.6</td><td>88.6</td><td>89.9</td><td>63.5</td><td>$0.0165 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">14.</span><td><a class="sys-name" href="https://github.com/allebee/jevk5" target="_blank" rel="noopener">JevK5 v0.2.0</a></td><td class="score">62.8</td><td>36.3</td><td>89.4</td><td>91.6</td><td>63.1</td><td>$0.017 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">15.</span><td><a class="sys-name" href="https://huggingface.co/HopitAI/hopper" target="_blank" rel="noopener">Hopper</a></td><td class="score">62.4</td><td>34.7</td><td>90.1</td><td>91.1</td><td>62.3</td><td>$0.0181 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">16.</span><td><a class="sys-name" href="https://github.com/fstandhartinger/jevbench/issues/110" target="_blank" rel="noopener">Imajev-4B (RTX 5090)</a></td><td class="score">61.9</td><td>34.6</td><td>89.2</td><td>90.5</td><td>63.3</td><td>$0.0168 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">17.</span><td><a class="sys-name" href="https://github.com/Octalab-Inc/jqv" target="_blank" rel="noopener">jqv (Qwen3-32B zero-shot)</a></td><td class="score">61.9</td><td>33.9</td><td>89.9</td><td>83.1</td><td>51.2</td><td>$0.0424 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">18.</span><td><a class="sys-name" href="https://github.com/metask-ai/metask-jev" target="_blank" rel="noopener">metask-jev-4b</a></td><td class="score">61.8</td><td>39.2</td><td>84.4</td><td>90.8</td><td>57.7</td><td>$0.0256 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">19.</span><td><a class="sys-name" href="https://huggingface.co/interfaze-ai/lev" target="_blank" rel="noopener">lev (Interfaze AI, Qwen3.5-4B + LoRA r32, label-token readout + candidate-path head, per-bucket temperatures)</a></td><td class="score">61.7</td><td>41.1</td><td>82.3</td><td>88.0</td><td>61.8</td><td>$0.0188 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">20.</span><td><a class="sys-name" href="https://github.com/newfull5/malkuth" target="_blank" rel="noopener">Malkuth-4B (newfull5, Kev post-train)</a></td><td class="score">61.1</td><td>33.8</td><td>88.4</td><td>89.5</td><td>55.8</td><td>$0.0297 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">21.</span><td><a class="sys-name" href="https://benchmarkheaven.com/jev-models/decision-4b-v12" target="_blank" rel="noopener">Decision 4B v1.2 (FlyMyJev, Qwen3.5-4B + LoRA)</a></td><td class="score">60.4</td><td>32.4</td><td>88.4</td><td>93.9</td><td>63.1</td><td>$0.017 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">22.</span><td><a class="sys-name" href="https://huggingface.co/chinhnc/Quyet-1.0-Medium" target="_blank" rel="noopener">Quyet-1.0-Medium (Chinh Nguyen, Qwen3.5-4B decoder, option-letter logits)</a></td><td class="score">59.2</td><td>41.7</td><td>76.8</td><td>89.5</td><td>63.4</td><td>$0.0166 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">23.</span><td><a class="sys-name" href="https://github.com/nschlaepfer/manchego-serve" target="_blank" rel="noopener">Manchego v2.1</a></td><td class="score">58.3</td><td>32.5</td><td>84.0</td><td>90.7</td><td>64.3</td><td>$0.0155 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">24.</span><td><a class="sys-name" href="https://github.com/us/jev-local" target="_blank" rel="noopener">jev-local (Qwen3.5-9B)</a></td><td class="score">57.5</td><td>41.9</td><td>73.1</td><td>75.6</td><td>58.9</td><td>$0.0235 <span class="est">est.</span></td><td>0.4</td></tr>
<tr><span class="rank">25.</span><td><a class="sys-name" href="https://benchmarkheaven.com/jev-models/decision-4b-v11" target="_blank" rel="noopener">Decision 4B v1.1 (FlyMyJev, Qwen3.5-4B + LoRA)</a></td><td class="score">57.2</td><td>29.1</td><td>85.2</td><td>94.0</td><td>63.1</td><td>$0.017 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">26.</span><td><a class="sys-name" href="https://huggingface.co/WIlfLin/JEV-Qwen3.5-9B-Base-NVFP4" target="_blank" rel="noopener">JEV Qwen3.5-9B Base NVFP4</a></td><td class="score">57.1</td><td>29.3</td><td>85.0</td><td>93.6</td><td>47.6</td><td>$0.0558 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">27.</span><td><a class="sys-name" href="https://github.com/mihail-gribov/typecastlm" target="_blank" rel="noopener">typecastlm (Mikhail Gribov, Qwen3.5-4B computed head)</a></td><td class="score">56.5</td><td>29.8</td><td>83.3</td><td>92.8</td><td>64.0</td><td>$0.0159 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">28.</span><td><a class="sys-name" href="https://github.com/TheoLeeCJ/openjev" target="_blank" rel="noopener">SemIf, formerly OpenJev (Qwen3.5-4B, TheoLeeCJ)</a></td><td class="score">55.4</td><td>27.1</td><td>83.6</td><td>90.8</td><td>63.1</td><td>$0.017 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">29.</span><td><a class="sys-name" href="https://github.com/amithgc/local-jev" target="_blank" rel="noopener">local-jev Qwen3.5-4B</a></td><td class="score">55.0</td><td>24.1</td><td>85.8</td><td>86.1</td><td>59.3</td><td>$0.0228 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">30.</span><td><a class="sys-name" href="https://huggingface.co/flymy-ai/decision-2b-preview" target="_blank" rel="noopener">Decision 2B (FlyMy.AI, v59)</a></td><td class="score">54.8</td><td>22.9</td><td>86.8</td><td>90.4</td><td>66.4</td><td>$0.0132 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">31.</span><td><a class="sys-name" href="https://github.com/abhishek085/open-spark-jev" target="_blank" rel="noopener">spark-s1-4b-v6 (Open Spark Jev, abhishek085)</a></td><td class="score">54.0</td><td>43.2</td><td>64.8</td><td>87.0</td><td>60.4</td><td>$0.0209 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">32.</span><td><a class="sys-name" href="https://huggingface.co/zeroentropy/zerank-2-reranker" target="_blank" rel="noopener">ZeroEntropy zerank-2</a></td><td class="score">49.4</td><td>15.6</td><td>83.2</td><td>82.9</td><td>48.4</td><td>$0.0525 <span class="est">est.</span></td><td>0.2</td></tr>
<tr><span class="rank">33.</span><td><a class="sys-name" href="https://github.com/ikermoel/open-alternative-jev" target="_blank" rel="noopener">open-alternative-jev (Qwen3.5-4B, IkerMoel)</a></td><td class="score">49.1</td><td>22.3</td><td>75.9</td><td>91.3</td><td>63.3</td><td>$0.0168 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">34.</span><td><a class="sys-name" href="https://github.com/sabeel111/OpenSourceJev" target="_blank" rel="noopener">OpenSourceJev (Qwen3.5-4B Q4_K_M, native llama.cpp)</a></td><td class="score">48.6</td><td>23.2</td><td>73.9</td><td>77.1</td><td>69.1</td><td>$0.0107 <span class="est">est.</span></td><td>0.3</td></tr>
<tr><span class="rank">35.</span><td><a class="sys-name" href="https://github.com/newfull5/malkuth" target="_blank" rel="noopener">Malkuth-2B (newfull5, Kev post-train)</a></td><td class="score">47.6</td><td>15.9</td><td>79.3</td><td>92.0</td><td>65.6</td><td>$0.0141 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">36.</span><td><a class="sys-name" href="https://github.com/pst2154/Nemotron_Jev" target="_blank" rel="noopener">Nemotron Diffusion 8B (pst2154, optimized vLLM)</a></td><td class="score">47.5</td><td>20.4</td><td>74.5</td><td>93.8</td><td>55.5</td><td>$0.0305 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">37.</span><td><a class="sys-name" href="https://huggingface.co/Qwen/Qwen3-Reranker-4B" target="_blank" rel="noopener">Qwen3-Reranker-4B</a></td><td class="score">44.7</td><td>16.3</td><td>73.2</td><td>82.1</td><td>48.4</td><td>$0.0525 <span class="est">est.</span></td><td>0.2</td></tr>
<tr><span class="rank">38.</span><td><a class="sys-name" href="https://huggingface.co/Mapika/decider-2b" target="_blank" rel="noopener">decider-2b (Mapika)</a></td><td class="score">44.6</td><td>20.7</td><td>68.5</td><td>94.6</td><td>64.9</td><td>$0.0148 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">39.</span><td><a class="sys-name" href="https://huggingface.co/BAAI/bge-reranker-v2-m3" target="_blank" rel="noopener">BAAI bge-reranker-v2-m3</a></td><td class="score">44.1</td><td>0.1</td><td>88.1</td><td>93.9</td><td>59.3</td><td>$0.0227 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">40.</span><td><a class="sys-name" href="https://github.com/jaredpalmer/kev" target="_blank" rel="noopener">kev 4B (research preview)</a></td><td class="score">44.0</td><td>19.5</td><td>68.5</td><td>90.5</td><td>65.8</td><td>$0.0138 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">41.</span><td><a class="sys-name" href="https://huggingface.co/mixedbread-ai/mxbai-rerank-base-v2" target="_blank" rel="noopener">Mixedbread mxbai-rerank-base-v2</a></td><td class="score">43.9</td><td>1.1</td><td>86.6</td><td>92.6</td><td>60.3</td><td>$0.021 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">42.</span><td><a class="sys-name" href="https://github.com/franckverrot/lev" target="_blank" rel="noopener">lev-350m (Franck Verrot, LFM2.5-350M)</a></td><td class="score">43.3</td><td>4.8</td><td>81.9</td><td>94.0</td><td>80.0</td><td>$0.00463 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">43.</span><td><a class="sys-name" href="https://huggingface.co/altslate/certo-decision-model" target="_blank" rel="noopener">Certo v1 (AltSlate Labs)</a></td><td class="score">43.2</td><td>0.2</td><td>86.1</td><td>92.5</td><td>96.9</td><td>$0.00127 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">44.</span><td><a class="sys-name" href="https://github.com/isHeSatoshi/smalljev" target="_blank" rel="noopener">smalljev semantic-v9</a></td><td class="score">43.0</td><td>8.7</td><td>77.2</td><td>90.2</td><td>60.8</td><td>$0.0203 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">45.</span><td><a class="sys-name" href="https://huggingface.co/Alibaba-NLP/gte-reranker-modernbert-base" target="_blank" rel="noopener">Alibaba GTE Reranker ModernBERT-base</a></td><td class="score">41.7</td><td>0.0</td><td>83.5</td><td>93.8</td><td>69.3</td><td>$0.0106 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">46.</span><td><a class="sys-name" href="https://huggingface.co/flymy-ai/decision-fast-preview" target="_blank" rel="noopener">Decision Fast (FlyMy.AI, v53a)</a></td><td class="score">40.6</td><td>6.1</td><td>75.1</td><td>91.5</td><td>80.1</td><td>$0.00461 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">47.</span><td><a class="sys-name" href="https://huggingface.co/heman10x/rlcd-modernbert-151m" target="_blank" rel="noopener">openJev Verdict 1.4</a></td><td class="score">40.4</td><td>5.0</td><td>75.7</td><td>81.5</td><td>86.6</td><td>$0.00279 <span class="est">est.</span></td><td>0.3</td></tr>
<tr><span class="rank">48.</span><td><a class="sys-name" href="https://huggingface.co/chinhnc/Quyet-1.0-Small-EN" target="_blank" rel="noopener">Quyet-1.0-Small-EN (Chinh Nguyen, ModernBERT-base encoder with fixed heads)</a></td><td class="score">39.5</td><td>6.1</td><td>73.0</td><td>94.9</td><td>89.1</td><td>$0.00231 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">49.</span><td><a class="sys-name" href="https://github.com/deepanwadhwa/OpenDecision" target="_blank" rel="noopener">OpenDecision (ModernBERT-large zero-shot)</a></td><td class="score">38.0</td><td>7.4</td><td>68.6</td><td>87.1</td><td>79.1</td><td>$0.00497 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">50.</span><td><a class="sys-name" href="https://github.com/Manavarya09/verdict" target="_blank" rel="noopener">verdict-small (Manavarya09, multilingual-e5-small 118M)</a></td><td class="score">37.6</td><td>2.3</td><td>72.9</td><td>89.5</td><td>100.0</td><td>$0.000866 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">51.</span><td><a class="sys-name" href="https://github.com/jaredpalmer/kev" target="_blank" rel="noopener">kev 0.6B (research preview)</a></td><td class="score">36.9</td><td>7.6</td><td>66.3</td><td>91.5</td><td>80.1</td><td>$0.00461 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">52.</span><td><a class="sys-name" href="https://huggingface.co/microsoft/Phi-4-mini-instruct" target="_blank" rel="noopener">Raw Phi-4 mini direct logits</a></td><td class="score">35.3</td><td>13.7</td><td>57.0</td><td>91.9</td><td>53.2</td><td>$0.0363 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">53.</span><td><a class="sys-name" href="https://fastino.ai/" target="_blank" rel="noopener">Fastino GLiNER-2.5-Decide (hosted API)</a></td><td class="score">34.6</td><td>7.0</td><td>62.1</td><td>84.7</td><td>45.8</td><td>$0.064</td><td>0.4</td></tr>
<tr><span class="rank">54.</span><td><a class="sys-name" href="https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507" target="_blank" rel="noopener">Raw Qwen3 4B Instruct 2507 direct logits</a></td><td class="score">34.6</td><td>34.0</td><td>35.2</td><td>91.2</td><td>63.4</td><td>$0.0166 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">55.</span><td><a class="sys-name" href="https://github.com/jaredpalmer/kev" target="_blank" rel="noopener">kev 0.5B</a></td><td class="score">32.9</td><td>4.6</td><td>61.2</td><td>92.4</td><td>80.1</td><td>$0.00461 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">56.</span><td><a class="sys-name" href="https://huggingface.co/chinhnc/Quyet-1.0-Small" target="_blank" rel="noopener">Quyet-1.0-Small (Chinh Nguyen, SEA-LION-ModernBERT-300M encoder with fixed heads)</a></td><td class="score">32.5</td><td>4.0</td><td>61.0</td><td>94.9</td><td>79.6</td><td>$0.0048 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">57.</span><td><a class="sys-name" href="https://huggingface.co/chinhnc/Quyet-1.0-Tiny" target="_blank" rel="noopener">Quyet-1.0-Tiny (Chinh Nguyen, mmBERT-small encoder (16 layers kept) with fixed heads)</a></td><td class="score">31.1</td><td>3.7</td><td>58.5</td><td>95.1</td><td>88.4</td><td>$0.00243 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">58.</span><td><a class="sys-name" href="https://github.com/JoshuaSP/open-jev" target="_blank" rel="noopener">Open Jev JSON Canvas (JoshuaSP)</a></td><td class="score">30.6</td><td>61.2</td><td>0.0</td><td>85.5</td><td>49.3</td><td>$0.049 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="rank">59.</span><td><a class="sys-name" href="https://github.com/Heman10x-NGU/openJev-verdict-2.0" target="_blank" rel="noopener">openJev Verdict (heman10x, ModernBERT-base 151M)</a></td><td class="score">26.6</td><td>4.1</td><td>49.1</td><td>79.4</td><td>86.6</td><td>$0.00279 <span class="est">est.</span></td><td>0.3</td></tr>
<tr><span class="rank">60.</span><td><a class="sys-name" href="https://github.com/Contrastive-LM/CLM" target="_blank" rel="noopener">CLM-8B (Contrastive-LM, clm-latest)</a></td><td class="score">20.1</td><td>0.1</td><td>40.0</td><td>92.6</td><td>50.5</td><td>$0.0447 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">61.</span><td><a class="sys-name" href="https://huggingface.co/LibertAIDAI/deem-0.8-v1" target="_blank" rel="noopener">Deem 0.8B v1</a></td><td class="score">18.1</td><td>6.5</td><td>29.6</td><td>85.4</td><td>80.3</td><td>$0.00454 <span class="est">est.</span></td><td>0.2</td></tr>
<tr><span class="rank">62.</span><td><a class="sys-name" href="https://huggingface.co/convaiinnovations/laya/tree/1c5edc17a7acd8701df6fc341c0d179f1c62c982/multilingual" target="_blank" rel="noopener">Laya multilingual</a></td><td class="score">17.0</td><td>0.3</td><td>33.7</td><td>77.8</td><td>82.2</td><td>$0.00393 <span class="est">est.</span></td><td>0.4</td></tr>
<tr><span class="rank">63.</span><td><a class="sys-name" href="https://huggingface.co/Qwen/Qwen3-1.7B" target="_blank" rel="noopener">Raw Qwen3 1.7B direct logits</a></td><td class="score">8.3</td><td>5.1</td><td>11.5</td><td>93.2</td><td>68.5</td><td>$0.0112 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="rank">64.</span><td><a class="sys-name" href="https://huggingface.co/Qwen/Qwen3-0.6B" target="_blank" rel="noopener">Raw Qwen3 0.6B direct logits</a></td><td class="score">8.2</td><td>5.6</td><td>10.9</td><td>93.5</td><td>77.6</td><td>$0.00559 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="latency 2.6× Jev (v1.5 reference)"<td><a class="sys-name" href="https://wity.alphanimble.com/" target="_blank" rel="noopener">wity-1 (Wity, reasoning auto)</a></td><td class="score">79.1</td><td>70.3</td><td>87.9</td><td>71.8</td><td>58.8</td><td>$0.0236</td><td>1.6</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 2.6× Jev (v1.5 reference)"<td><a class="sys-name" href="https://github.com/blockbrain-ai/swanone-recipe" target="_blank" rel="noopener">swanOne (blockbrain, Qwen3.8-Flash-Next NVFP4)</a></td><td class="score">72.4</td><td>56.2</td><td>88.6</td><td>82.6</td><td>42.2</td><td>$0.0848 <span class="est">est.</span></td><td>0.3</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 2.5× Jev (v1.5 reference)"<td><a class="sys-name" href="https://github.com/igorls/ninfer" target="_blank" rel="noopener">NInfer Qwen3.8-Flash-Next mixed</a></td><td class="score">69.7</td><td>49.0</td><td>90.4</td><td>89.3</td><td>42.5</td><td>$0.0823 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 2.0× Jev (v1.5 reference)"<td><a class="sys-name" href="https://huggingface.co/jsaurabh/qwen3.5-9b-jev-data-mix-v2" target="_blank" rel="noopener">Qwen3.5-9B Jev-like data-mix v2</a></td><td class="score">60.9</td><td>41.9</td><td>79.8</td><td>85.4</td><td>45.7</td><td>$0.0646 <span class="est">est.</span></td><td>0.2</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 3.1× Jev (v1.5 reference)"<td><a class="sys-name" href="https://huggingface.co/juspay/jev-one" target="_blank" rel="noopener">JevOne</a></td><td class="score">68.9</td><td>46.0</td><td>91.8</td><td>89.8</td><td>39.8</td><td>$0.101 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 4.8× Jev (v1.5 reference)"<td><a class="sys-name" href="https://huggingface.co/Mapika/decider-35b-a3b" target="_blank" rel="noopener">decider-35b-a3b (Mapika)</a></td><td class="score">66.2</td><td>48.7</td><td>83.6</td><td>91.3</td><td>34.4</td><td>$0.154 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 4.0× Jev (v1.5 reference)"<td><a class="sys-name" href="https://github.com/bespokelabsai/nimble" target="_blank" rel="noopener">Bespoke Nimble 9B (Bespoke Labs)</a></td><td class="score">56.5</td><td>43.1</td><td>69.9</td><td>86.1</td><td>36.8</td><td>$0.128 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="latency 4.6× Jev (v1.5 reference)"<td><a class="sys-name" href="https://github.com/kshetrajna12/reflex" target="_blank" rel="noopener">reflex 4B (kshetrajna12)</a></td><td class="score">59.0</td><td>30.9</td><td>87.0</td><td>68.8</td><td>63.1</td><td>$0.0169 <span class="est">est.</span></td><td>1.4</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 7.0× Jev (v1.5 reference)"<td><a class="sys-name" href="https://huggingface.co/denis-pplx/autojev-27b" target="_blank" rel="noopener">AutoJev-27B (denis-pplx, Qwen3.8-27B)</a></td><td class="score">71.6</td><td>55.6</td><td>87.6</td><td>88.6</td><td>29.4</td><td>$0.226 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 7.4× Jev (v1.5 reference)"<td><a class="sys-name" href="https://huggingface.co/caiovicentino1/eikos-27b" target="_blank" rel="noopener">Eikos-27B (caiovicentino1, Qwen3.8-27B)</a></td><td class="score">77.2</td><td>64.4</td><td>90.0</td><td>88.1</td><td>28.7</td><td>$0.238 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 7.1× Jev (v1.5 reference)"<td><a class="sys-name" href="https://github.com/igorls/ninfer" target="_blank" rel="noopener">NInfer Qwen3.8-27B NVFP4</a></td><td class="score">68.0</td><td>51.8</td><td>84.2</td><td>90.5</td><td>29.1</td><td>$0.231 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 7.5× Jev (v1.5 reference); latency 3.0× Jev (v1.5 reference)"<td><a class="sys-name" href="https://github.com/razorback16/openjev" target="_blank" rel="noopener">OpenJev (thinking, BF16)</a></td><td class="score">75.9</td><td>70.3</td><td>81.4</td><td>72.6</td><td>28.5</td><td>$0.241 <span class="est">est.</span></td><td>0.9</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 5.3× Jev (v1.5 reference); latency 2.2× Jev (v1.5 reference)"<td><a class="sys-name" href="https://github.com/Zefan-Cai/Open-Jev" target="_blank" rel="noopener">Open-Jev 9B (Zefan Cai)</a></td><td class="score">61.8</td><td>43.9</td><td>79.8</td><td>74.4</td><td>33.1</td><td>$0.17 <span class="est">est.</span></td><td>0.6</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 2.4× Jev (v1.5 reference)"<td><a class="sys-name" href="https://huggingface.co/StandardThinking/StandardOne-8B" target="_blank" rel="noopener">Standard One 8B (Standard Thinking)</a></td><td class="score">58.4</td><td>32.8</td><td>84.0</td><td>92.8</td><td>43.3</td><td>$0.0777 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 7.7× Jev (v1.5 reference)"<td><a class="sys-name" href="https://huggingface.co/Cloudflare/clef" target="_blank" rel="noopener">Clef (Cloudflare, Qwen3.8-27B post-train with a joint schema head, multimodal, measured on text)</a></td><td class="score">75.3</td><td>59.1</td><td>91.6</td><td>83.4</td><td>28.1</td><td>$0.249 <span class="est">est.</span></td><td>0.2</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 7.6× Jev (v1.5 reference); latency 3.4× Jev (v1.5 reference)"<td><a class="sys-name" href="https://huggingface.co/Reza2kn/Bev" target="_blank" rel="noopener">Bev / Bonsai 27B</a></td><td class="score">66.1</td><td>43.8</td><td>88.5</td><td>73.0</td><td>28.2</td><td>$0.247 <span class="est">est.</span></td><td>1.0</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 7.6× Jev (v1.5 reference); latency 5.0× Jev (v1.5 reference)"<td><a class="sys-name" href="https://github.com/zhengxuyu/litjev" target="_blank" rel="noopener">LitJev (Qwen3.8-27B)</a></td><td class="score">66.6</td><td>43.3</td><td>89.9</td><td>68.2</td><td>28.4</td><td>$0.244 <span class="est">est.</span></td><td>1.5</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 2.1× Jev (v1.5 reference)"<td><a class="sys-name" href="https://github.com/sgoedecke/system-one" target="_blank" rel="noopener">system-one (Qwen3-8B, Sean Goedecke)</a></td><td class="score">29.7</td><td>29.3</td><td>30.2</td><td>90.9</td><td>45.0</td><td>$0.0683 <span class="est">est.</span></td><td>0.0</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 3.0× Jev (v1.5 reference)"<td><a class="sys-name" href="https://github.com/jaredpalmer/kev" target="_blank" rel="noopener">kev 8B (research preview)</a></td><td class="score">50.0</td><td>29.3</td><td>70.8</td><td>89.0</td><td>40.4</td><td>$0.0968 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 2.0× Jev (v1.5 reference)"<td><a class="sys-name" href="https://huggingface.co/Qwen/Qwen3-8B" target="_blank" rel="noopener">Raw Qwen3 8B direct logits</a></td><td class="score">30.1</td><td>26.5</td><td>33.7</td><td>88.7</td><td>45.5</td><td>$0.0654 <span class="est">est.</span></td><td>0.1</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 5.3× Jev (v1.5 reference)"<td><a class="sys-name" href="https://github.com/Zefan-Cai/Open-Jev" target="_blank" rel="noopener">Open-Jev 2B (Zefan Cai)</a></td><td class="score">47.7</td><td>21.8</td><td>73.6</td><td>77.1</td><td>33.1</td><td>$0.17 <span class="est">est.</span></td><td>0.4</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="cost 22.2× Jev (v1.5 reference); latency 2.8× Jev (v1.5 reference)"<td><a class="sys-name" href="https://huggingface.co/ZefanCai/Open-Jev-27B-v1.1" target="_blank" rel="noopener">Open-Jev 27B v1.1 (Zefan Cai, LoRA + decision head on Qwen3.8-27B)</a></td><td class="score">68.0</td><td>52.5</td><td>83.5</td><td>72.5</td><td>14.4</td><td>$0.716 <span class="est">est.</span></td><td>0.8</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="latency 14.3× Jev (v1.5 reference)"<td><a class="sys-name" href="https://github.com/featherless-ai/simple-jev" target="_blank" rel="noopener">SimpleJev (Qwen3.5-0.8B, CPU)</a></td><td class="score">31.6</td><td>13.0</td><td>50.2</td><td>58.9</td><td>70.3</td><td>$0.00976 <span class="est">est.</span></td><td>4.3</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="latency 18.3× Jev (v1.5 reference)"<td><a class="sys-name" href="https://huggingface.co/mghafiri/qwen3.5-0.8B-decision-model" target="_blank" rel="noopener">Qwen3.5-0.8B Decision Model (Mourad Ghafiri)</a></td><td class="score">42.6</td><td>7.3</td><td>78.0</td><td>55.9</td><td>79.5</td><td>$0.00482 <span class="est">est.</span></td><td>5.6</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="latency 3.0× Jev (v1.5 reference)"<td><a class="sys-name" href="https://huggingface.co/convaiinnovations/laya/tree/1c5edc17a7acd8701df6fc341c0d179f1c62c982/typed-decisions" target="_blank" rel="noopener">Laya typed-decisions</a></td><td class="score">44.2</td><td>5.5</td><td>83.0</td><td>70.8</td><td>82.5</td><td>$0.00382 <span class="est">est.</span></td><td>0.9</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="latency 5.7× Jev (v1.5 reference)"<td><a class="sys-name" href="https://github.com/kotoba-lang/typed-decisions" target="_blank" rel="noopener">open-jev-deberta-v3-large (local CPU)</a></td><td class="score">35.4</td><td>2.8</td><td>68.0</td><td>67.6</td><td>77.6</td><td>$0.00558 <span class="est">est.</span></td><td>1.7</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="latency 2.9× Jev (v1.5 reference)"<td><a class="sys-name" href="https://huggingface.co/convaiinnovations/laya" target="_blank" rel="noopener">Laya (Convai Innovations, ModernBERT-large 421M)</a></td><td class="score">32.5</td><td>1.8</td><td>63.2</td><td>73.3</td><td>84.9</td><td>$0.00319 <span class="est">est.</span></td><td>0.8</td></tr>
<tr><span class="muted" aria-label="not within official caps">—</span> title="latency 2.4× Jev (v1.5 reference)"<td><a class="sys-name" href="https://github.com/Bluusun/Decision-API" target="_blank" rel="noopener">Mirror</a></td><td class="score">25.8</td><td>0.0</td><td>51.6</td><td>73.5</td><td>89.3</td><td>$0.00228 <span class="est">est.</span></td><td>0.7</td></tr>
</tbody>
</table>
</div>
<h2>Best Jev alternatives</h2>
<p class="lede" style="font-size:15px">The strongest other systems in this snapshot, by Capability Score: <a href="https://huggingface.co/chinhnc/Quyet-1.0-Large" target="_blank" rel="noopener">Quyet-1.0-Large</a> (81.7) · <a href="https://levanto.ai" target="_blank" rel="noopener">Sage 1.3.0</a> (78.6) · <a href="https://github.com/krishna-gogineni-765/deck31b" target="_blank" rel="noopener">deck-31B</a> (77.6). The Jev reference row (TypeSafe AI) is excluded from this list. For the full comparison — open-weight options, use-case fit and per-row evidence — see the <a href="https://benchmarkheaven.com/jev-models/alternatives" target="_blank" rel="noopener">Jev alternatives page</a>.</p>
<h2>How the score works</h2>
<div class="method">
<div class="method-card"><b>Intelligence</b><span>Chance-corrected accuracy on the frozen public and sealed tiers.</span></div>
<div class="method-card"><b>Calibration</b><span>Distribution quality of the typed answers a system produces.</span></div>
<div class="method-card"><b>Speed</b><span>Median (p50) latency on a logarithmic scale; self-hosted runs carry a published adjustment.</span></div>
<div class="method-card"><b>Cost</b><span>USD per 1,000 complete decisions — a whole question, not per 1,000 tokens.</span></div>
</div>
<p class="formula"><b>JevBench Capability Score:</b> Arithmetic mean of Intelligence and Calibration; ranked models within both official caps. The official caps are 2&times; the Jev reference (jev-1.13.0@v1.5.7): at most 0.0646 USD per 1,000 decisions — 1.23 s median latency. The <b>Composite</b> is a weighted harmonic mean of Intelligence, Calibration, Speed and Cost, with equal 25% weights (not an arithmetic average) and configured squared penalties; it is secondary on open weights and headlines the hosted API board. Cost figures are labeled by evidence basis; &ldquo;est.&rdquo; means estimated, not a measured charge.</p>
<h2>Listed, not ranked</h2>
<p class="note">These systems appear on the board but are not part of the ranking in this release.</p>
<ul class="listed">
<li><b>wity-1 (Wity, reasoning always)</b> — listed: <span class="muted">variant listed for reference, not ranked in this release</span></li>
<li><b>wity-1 (Wity, reasoning off)</b> — listed: <span class="muted">variant listed for reference, not ranked in this release</span></li>
</ul>
<h2>More</h2>
<ul class="links">
<li><a href="https://benchmarkheaven.com/jev-models" target="_blank" rel="noopener">Live JevBench board</a> — full interactive view, filters and head-to-heads</li>
<li><a href="https://benchmarkheaven.com/jev-models/alternatives" target="_blank" rel="noopener">Jev alternatives, compared</a></li>
<li><a href="https://benchmarkheaven.com/jev-models/how-to-choose" target="_blank" rel="noopener">How to choose a Jev-class model</a></li>
<li><a href="https://benchmarkheaven.com/jev-models/open-source-jev" target="_blank" rel="noopener">Is Jev open source?</a></li>
<li><a href="https://benchmarkheaven.com/api/jevbench/latest" target="_blank" rel="noopener">Public results API (JSON)</a></li>
<li><a href="https://benchmarkheaven.com/submit" target="_blank" rel="noopener">Submit a system to JevBench</a></li>
<li><a href="https://github.com/fstandhartinger/jevbench" target="_blank" rel="noopener">JevBench method repository (GitHub)</a></li>
<li><a href="https://huggingface.co/datasets/benchmarkheaven/jevbench" target="_blank" rel="noopener">JevBench results dataset (Hugging Face)</a></li>
</ul>
<footer class="foot">
JevBench is maintained by <a href="https://benchmarkheaven.com" target="_blank" rel="noopener">Benchmark Heaven</a>
and is not affiliated with or endorsed by TypeSafe AI; Jev is one of the systems evaluated, not the benchmark owner.
</footer>
</main>
<script>
(function () {
var root = document.documentElement;
var saved = localStorage.getItem('jevbench-theme');
if (saved) root.setAttribute('data-theme', saved);
else if (window.matchMedia && window.matchMedia('(prefers-color-scheme: light)').matches) root.setAttribute('data-theme', 'light');
document.getElementById('themeToggle').addEventListener('click', function () {
var next = root.getAttribute('data-theme') === 'dark' ? 'light' : 'dark';
root.setAttribute('data-theme', next);
localStorage.setItem('jevbench-theme', next);
});
})();
(function () {
var table = document.getElementById('board');
if (!table) return;
var tbody = table.tBodies[0];
table.querySelectorAll('thead th').forEach(function (th, idx) {
th.style.cursor = 'pointer';
th.title = 'Click to sort by this column';
th.addEventListener('click', function () {
var numeric = th.getAttribute('data-type') === 'num';
var dir = th.getAttribute('data-dir') === 'asc' ? -1 : 1;
table.querySelectorAll('thead th').forEach(function (o) { o.removeAttribute('data-dir'); });
th.setAttribute('data-dir', dir === 1 ? 'asc' : 'desc');
var rows = Array.prototype.slice.call(tbody.rows);
rows.sort(function (a, b) {
var av = a.cells[idx] ? a.cells[idx].textContent.trim() : '';
var bv = b.cells[idx] ? b.cells[idx].textContent.trim() : '';
if (numeric) {
var an = parseFloat(av.replace(/[^0-9.eE+-]/g, ''));
var bn = parseFloat(bv.replace(/[^0-9.eE+-]/g, ''));
if (!isNaN(an) && !isNaN(bn)) return (an - bn) * dir;
return (isNaN(an) - isNaN(bn)) * dir;
}
return av.localeCompare(bv) * dir;
});
rows.forEach(function (r) { tbody.appendChild(r); });
});
});
})();
</script>
</body>
</html>