Ergebnisse aktualisiert (12 Läufe)

app.js

@@ -116,68 +116,162 @@
     }
   };
 
-  var overview = readJson("overview-data");
-  if (overview && overview.bars.length) {
-    new Chart(document.getElementById("bar-chart"), {
-      type: "bar",
-      data: {
-        labels: overview.bars.map(function (b) { return b.label; }),
-        datasets: [{
-          label: "Gesamtscore",
-          data: overview.bars.map(function (b) { return b.value; }),
-          backgroundColor: series[0],
-          borderRadius: 4,
-          borderSkipped: "start",
-          barPercentage: 0.8,
-          categoryPercentage: 0.9,
-          maxBarThickness: 18
-        }]
-      },
-      options: {
-        indexAxis: "y",
-        maintainAspectRatio: false,
-        animation: false,
-        layout: { padding: { right: 40 } },
-        scales: { x: axis("Score (0–100)", { min: 0, max: 100 }), y: axis(null, { grid: { display: false } }) },
-        plugins: {
-          legend: { display: false },
-          tooltip: { callbacks: { label: function (c) { return "Gesamt: " + c.parsed.x.toFixed(1); } } }
-        }
-      },
-      plugins: [barValueLabels]
-    });
+  // Wunschbereich im Streudiagramm: rechts von min. Tokens/s und über min. Note, leicht grün hinterlegt.
+  // Fehlt einer der Werte, ist der Bereich zu dieser Seite hin offen.
+  var zone = token("--zone");
+  var zonePlugin = {
+    id: "zone",
+    beforeDatasetsDraw: function (chart, args, opts) {
+      if (!opts || (opts.minX == null && opts.minY == null)) return;
+      var area = chart.chartArea;
+      var left = opts.minX == null ? area.left : Math.max(area.left, chart.scales.x.getPixelForValue(opts.minX));
+      var bottom = opts.minY == null ? area.bottom : Math.min(area.bottom, chart.scales.y.getPixelForValue(opts.minY));
+      if (left >= area.right || bottom <= area.top) return;
+      var ctx = chart.ctx;
+      ctx.save();
+      ctx.fillStyle = zone;
+      ctx.fillRect(left, area.top, area.right - left, bottom - area.top);
+      ctx.restore();
+    }
+  };
+
+  // Gewählte Kategorie je Browser merken – nur Bequemlichkeit, ohne Speicher geht es mit „Gesamt“ los.
+  var CATEGORY_KEY = "bench.chartCategory";
+  function rememberedCategory() {
+    try { return window.localStorage.getItem(CATEGORY_KEY); } catch (e) { return null; }
+  }
+  function rememberCategory(key) {
+    try { window.localStorage.setItem(CATEGORY_KEY, key); } catch (e) { /* ohne Speicher */ }
   }
 
-  if (overview && overview.scatter.length) {
-    new Chart(document.getElementById("scatter-chart"), {
-      type: "scatter",
-      data: {
-        datasets: [{
-          label: "Modellvarianten",
-          data: overview.scatter,
-          backgroundColor: series[0],
-          borderColor: surface,
-          borderWidth: 2,
-          pointRadius: 5,
-          pointHoverRadius: 7,
-          pointHitRadius: 12
-        }]
-      },
-      options: {
-        maintainAspectRatio: false,
-        animation: false,
-        scales: { x: axis("Generierung (Tokens/s)", { beginAtZero: true }), y: axis("Gesamtscore", { min: 0, max: 100 }) },
-        plugins: {
-          legend: { display: false },
-          tooltip: {
-            callbacks: {
-              label: function (c) { return c.raw.label + ": " + c.raw.y.toFixed(1) + " Punkte, " + c.raw.x.toFixed(1) + " t/s"; }
+  // Übersicht (Webseite und Space): je .chart-set ein Balken- und ein Streudiagramm, umschaltbar nach Kategorie.
+  document.querySelectorAll(".chart-set").forEach(function (set) {
+    var dataEl = set.querySelector("script.chart-data");
+    if (!dataEl) return;
+    var data = JSON.parse(dataEl.textContent);
+    var texts = data.texts;
+    var limits = data.thresholds || {};
+    var buttons = Array.prototype.slice.call(set.querySelectorAll("[data-chart-category]"));
+    var barBox = set.querySelector('[data-chart-box="bar"]');
+    var barTitle = set.querySelector('[data-chart-title="bar"]');
+    var scatterTitle = set.querySelector('[data-chart-title="scatter"]');
+    var barChart = null, scatterChart = null;
+
+    function titleOf(key) {
+      if (key === "overall") return texts.overall_title;
+      var match = data.categories.filter(function (c) { return c.key === key; })[0];
+      return match ? match.label : key;
+    }
+
+    function show(key) {
+      var title = titleOf(key);
+      buttons.forEach(function (b) {
+        var picked = b.dataset.chartCategory === key;
+        b.classList.toggle("active", picked);
+        b.setAttribute("aria-pressed", picked ? "true" : "false");
+      });
+      if (barTitle) barTitle.textContent = title;
+      if (scatterTitle) scatterTitle.textContent = title + texts.vs;
+      var bars = data.runs.filter(function (r) { return r.scores[key] !== null && r.scores[key] !== undefined; })
+        .sort(function (a, b) { return b.scores[key] - a.scores[key]; });
+      var points = bars.filter(function (r) { return r.tps !== null; })
+        .map(function (r) { return { label: r.label, x: r.tps, y: r.scores[key], source: r.tps_source }; });
+      // Hohl: Geschwindigkeit nur aus dem Speed-Test, der Lauf lieferte keine Werte.
+      var fills = points.map(function (p) { return p.source === "test" ? surface : series[0]; });
+      var edges = points.map(function (p) { return p.source === "test" ? series[0] : surface; });
+
+      // Balken der Modelle im Wunschbereich grün – dieselbe Regel wie das Rechteck im Streudiagramm.
+      var minTps = limits.min_gen_tps == null ? null : limits.min_gen_tps;
+      var minScore = limits.min_score == null ? null : limits.min_score;
+      function inZone(r) {
+        if (minTps === null && minScore === null) return false;
+        if (minTps !== null && (r.tps === null || r.tps < minTps)) return false;
+        return minScore === null || r.scores[key] >= minScore;
+      }
+
+      if (barChart) barChart.destroy();
+      if (barBox) barBox.style.height = (70 + 30 * Math.max(bars.length, 1)) + "px";
+      barChart = new Chart(set.querySelector('[data-chart="bar"]'), {
+        type: "bar",
+        data: {
+          labels: bars.map(function (r) { return r.label; }),
+          datasets: [{
+            label: title,
+            data: bars.map(function (r) { return r.scores[key]; }),
+            backgroundColor: bars.map(function (r) { return inZone(r) ? series[2] : series[0]; }),
+            borderRadius: 4,
+            borderSkipped: "start",
+            barPercentage: 0.8,
+            categoryPercentage: 0.9,
+            maxBarThickness: 18
+          }]
+        },
+        options: {
+          indexAxis: "y",
+          maintainAspectRatio: false,
+          animation: false,
+          layout: { padding: { right: 40 } },
+          scales: { x: axis(texts.score_axis, { min: 0, max: 100 }), y: axis(null, { grid: { display: false } }) },
+          plugins: {
+            legend: { display: false },
+            tooltip: { callbacks: { label: function (c) { return title + ": " + c.parsed.x.toFixed(1); } } }
+          }
+        },
+        plugins: [barValueLabels]
+      });
+
+      if (scatterChart) scatterChart.destroy();
+      var minX = limits.min_gen_tps == null ? null : limits.min_gen_tps;
+      var minY = limits.min_score == null ? null : limits.min_score;
+      scatterChart = new Chart(set.querySelector('[data-chart="scatter"]'), {
+        type: "scatter",
+        data: {
+          datasets: [{
+            label: title,
+            data: points,
+            backgroundColor: fills,
+            borderColor: edges,
+            borderWidth: 2,
+            pointRadius: 5,
+            pointHoverRadius: 7,
+            pointHitRadius: 12
+          }]
+        },
+        options: {
+          maintainAspectRatio: false,
+          animation: false,
+          scales: {
+            // Der Wunschbereich bleibt sichtbar, auch wenn alle Modelle langsamer sind.
+            x: axis(texts.tps_axis, { beginAtZero: true, suggestedMax: minX === null ? undefined : minX * 1.1 }),
+            y: axis(title, { min: 0, max: 100 })
+          },
+          plugins: {
+            legend: { display: false },
+            zone: { minX: minX, minY: minY },
+            tooltip: {
+              callbacks: {
+                label: function (c) {
+                  var source = c.raw.source === "test" ? texts.source_test : texts.source_run;
+                  return c.raw.label + ": " + c.raw.y.toFixed(1) + " " + texts.points + ", " + c.raw.x.toFixed(1) + " t/s (" + source + ")";
+                }
+              }
             }
           }
-        }
-      }
+        },
+        plugins: [zonePlugin]
+      });
+    }
+
+    buttons.forEach(function (b) {
+      b.addEventListener("click", function () {
+        show(b.dataset.chartCategory);
+        rememberCategory(b.dataset.chartCategory);
+      });
     });
-  }
+    var start = rememberedCategory();
+    var known = data.categories.some(function (c) { return c.key === start; });
+    show(known ? start : "overall");
+  });
 
   var categories = readJson("category-data");
   if (categories && categories.series.length) {

index.html

@@ -29,7 +29,7 @@
 <thead><tr>
   <th data-sort="text">Model</th>
   <th data-sort="num" class="num">Overall</th>
-<th data-sort="num" class="num">Knowledge &amp; reasoning</th><th data-sort="num" class="num">Math</th><th data-sort="num" class="num">Instruction following</th><th data-sort="num" class="num">Code</th><th data-sort="num" class="num">German</th><th data-sort="num" class="num">gen t/s</th><th data-sort="num" class="num">pp t/s 4k</th><th data-sort="num" class="num">TTFT 4k ms</th>  <th data-sort="text">Date</th>
+<th data-sort="num" class="num">Knowledge &amp; reasoning</th><th data-sort="num" class="num">Math</th><th data-sort="num" class="num">Instruction following</th><th data-sort="num" class="num">Code</th><th data-sort="num" class="num">German</th><th data-sort="num" class="num">gen t/s</th><th data-sort="num" class="num" title="Generation per request measured during the benchmark run; gen t/s before it comes from the synthetic speed test">gen t/s run</th><th data-sort="num" class="num">pp t/s 4k</th><th data-sort="num" class="num">TTFT 4k ms</th>  <th data-sort="text">Date</th>
 </tr></thead>
 <tbody>
 <tr data-reasoning="on">
@@ -44,6 +44,7 @@
   <td class="num strong" data-value="83.0" title="livecodebench_gen: 83.0 ± 2.7 (n=200)&#10;">83.0</td>
   <td class="num" data-value="77.36507936507935" title="mmlu_prox_de: 80.3 ± 2.3 (n=300)&#10;include_de_gen: 71.4 ± 3.7 (n=154)&#10;">77.4</td>
   <td class="num" data-value="">–</td>
+  <td class="num" data-value="19.638176195686693" title="overall 58.4 t/s with 3 parallel requests, from 9 of 9 tasks">20</td>
   <td class="num" data-value="">–</td>
   <td class="num" data-value="">–</td>
   <td data-value="2026-09-18">2026-09-18</td>
@@ -60,6 +61,7 @@
   <td class="num" data-value="75.5" title="livecodebench_gen: 75.5 ± 3.0 (n=200)&#10;">75.5</td>
   <td class="num strong" data-value="81.13708513708514" title="mmlu_prox_de: 85.7 ± 2.0 (n=300)&#10;include_de_gen: 72.1 ± 3.6 (n=154)&#10;">81.1</td>
   <td class="num" data-value="26.376464212349354">26</td>
+  <td class="num" data-value="21.59310838133867" title="overall 64.4 t/s with 3 parallel requests, from 5 of 9 tasks">22</td>
   <td class="num" data-value="2041.4640201348918">2041</td>
   <td class="num" data-value="1973.583644023165">1974</td>
   <td data-value="2026-09-17">2026-09-17</td>
@@ -76,6 +78,7 @@
   <td class="num" data-value="75.0" title="livecodebench_gen: 75.0 ± 3.1 (n=200)&#10;">75.0</td>
   <td class="num" data-value="74.51659451659451" title="mmlu_prox_de: 78.3 ± 2.4 (n=300)&#10;include_de_gen: 66.9 ± 3.8 (n=154)&#10;">74.5</td>
   <td class="num" data-value="16.51787775562755">17</td>
+  <td class="num" data-value="13.757625586585043" title="overall 41.2 t/s with 3 parallel requests, from 5 of 9 tasks">14</td>
   <td class="num" data-value="1359.0489942173106">1359</td>
   <td class="num" data-value="2952.064286917448">2952</td>
   <td data-value="2026-09-18">2026-09-18</td>
@@ -92,6 +95,7 @@
   <td class="num" data-value="75.0" title="livecodebench_gen: 75.0 ± 3.1 (n=200)&#10;">75.0</td>
   <td class="num" data-value="76.4935064935065" title="mmlu_prox_de: 80.0 ± 2.3 (n=300)&#10;include_de_gen: 69.5 ± 3.7 (n=154)&#10;">76.5</td>
   <td class="num" data-value="16.51271607105954">17</td>
+  <td class="num" data-value="13.819760405181762" title="overall 41.0 t/s with 3 parallel requests, from 5 of 9 tasks">14</td>
   <td class="num" data-value="1358.107397713914">1358</td>
   <td class="num" data-value="2954.110997961834">2954</td>
   <td data-value="2026-09-19">2026-09-19</td>
@@ -108,10 +112,28 @@
   <td class="num" data-value="65.0" title="livecodebench_gen: 65.0 ± 3.4 (n=200)&#10;">65.0</td>
   <td class="num" data-value="76.25974025974027" title="mmlu_prox_de: 79.0 ± 2.4 (n=300)&#10;include_de_gen: 70.8 ± 3.7 (n=154)&#10;">76.3</td>
   <td class="num" data-value="15.103805505442727">15</td>
+  <td class="num" data-value="22.383948355383822" title="overall 67.6 t/s with 3 parallel requests, from 5 of 9 tasks">22</td>
   <td class="num" data-value="2550.5619640606633">2551</td>
   <td class="num" data-value="1577.6915270835161">1578</td>
   <td data-value="2026-09-18">2026-09-18</td>
 </tr>
+<tr data-reasoning="off">
+  <td data-value="occamy-1.0">
+    <a href="runs/27.html" class="strong">occamy-1.0</a>
+    <div class="sub">NVFP4 · Reasoning off · direct · NVIDIA DGX Spark, 128 GB RAM</div>
+  </td>
+  <td class="num strong" data-value="64.03163768005874">64.0</td>
+  <td class="num" data-value="73.1057239057239" title="mmlu_pro: 84.0 ± 2.1 (n=300)&#10;gpqa_diamond_gen: 71.7 ± 3.2 (n=198)&#10;musr_gen: 63.6 ± 3.0 (n=250)&#10;">73.1</td>
+  <td class="num" data-value="62.4" title="omni_math_gen: 62.4 ± 3.1 (n=250)&#10;">62.4</td>
+  <td class="num" data-value="44.73182957393484" title="ifbench: 31.3 ± 2.7 (n=300)&#10;ifeval: 85.3 ± 2.9 (n=150)&#10;">44.7</td>
+  <td class="num" data-value="63.0" title="livecodebench_gen: 63.0 ± 3.4 (n=200)&#10;">63.0</td>
+  <td class="num" data-value="76.92063492063492" title="mmlu_prox_de: 79.7 ± 2.3 (n=300)&#10;include_de_gen: 71.4 ± 3.7 (n=154)&#10;">76.9</td>
+  <td class="num" data-value="36.719412085128695">37</td>
+  <td class="num" data-value="31.562617796089718" title="overall 94.1 t/s with 3 parallel requests, from 9 of 9 tasks">32</td>
+  <td class="num" data-value="5539.414736774487">5539</td>
+  <td class="num" data-value="726.4305330463685">726</td>
+  <td data-value="2026-09-21">2026-09-21</td>
+</tr>
 <tr data-reasoning="off">
   <td data-value="qwen3.8-27b-abliterated">
     <a href="runs/23.html" class="strong">qwen3.8-27b-abliterated</a>
@@ -124,10 +146,28 @@
   <td class="num" data-value="59.5" title="livecodebench_gen: 59.5 ± 3.5 (n=200)&#10;">59.5</td>
   <td class="num" data-value="72.06637806637806" title="mmlu_prox_de: 74.3 ± 2.5 (n=300)&#10;include_de_gen: 67.5 ± 3.8 (n=154)&#10;">72.1</td>
   <td class="num" data-value="15.903617023850842">16</td>
+  <td class="num" data-value="21.942627395574263" title="overall 67.0 t/s with 3 parallel requests, from 9 of 9 tasks">22</td>
   <td class="num" data-value="2472.0116600865786">2472</td>
   <td class="num" data-value="1627.824036986567">1628</td>
   <td data-value="2026-09-20">2026-09-20</td>
 </tr>
+<tr data-reasoning="off">
+  <td data-value="ternary-bonsai-2-27b">
+    <a href="runs/26.html" class="strong">ternary-bonsai-2-27b</a>
+    <div class="sub">FP8 · Reasoning off · direct · NVIDIA DGX Spark, 128 GB RAM</div>
+  </td>
+  <td class="num strong" data-value="61.28656742867268">61.3</td>
+  <td class="num" data-value="69.5077441077441" title="mmlu_pro: 77.7 ± 2.4 (n=300)&#10;gpqa_diamond_gen: 65.7 ± 3.4 (n=198)&#10;musr_gen: 65.2 ± 3.0 (n=250)&#10;">69.5</td>
+  <td class="num" data-value="59.599999999999994" title="omni_math_gen: 59.6 ± 3.1 (n=250)&#10;">59.6</td>
+  <td class="num" data-value="46.32581453634085" title="ifbench: 34.3 ± 2.7 (n=300)&#10;ifeval: 82.7 ± 3.1 (n=150)&#10;">46.3</td>
+  <td class="num" data-value="56.49999999999999" title="livecodebench_gen: 56.5 ± 3.5 (n=200)&#10;">56.5</td>
+  <td class="num" data-value="74.4992784992785" title="mmlu_prox_de: 77.3 ± 2.4 (n=300)&#10;include_de_gen: 68.8 ± 3.7 (n=154)&#10;">74.5</td>
+  <td class="num" data-value="28.488199681723682">28</td>
+  <td class="num" data-value="19.303084023060574" title="overall 58.5 t/s with 3 parallel requests, from 9 of 9 tasks">19</td>
+  <td class="num" data-value="898.1292759650371">898</td>
+  <td class="num" data-value="4480.424041044898">4480</td>
+  <td data-value="2026-09-21">2026-09-21</td>
+</tr>
 <tr data-reasoning="off">
   <td data-value="gemma-4-12b-it-heretic">
     <a href="runs/17.html" class="strong">gemma-4-12b-it-heretic</a>
@@ -140,6 +180,7 @@
   <td class="num" data-value="62.0" title="livecodebench_gen: 62.0 ± 3.4 (n=200)&#10;">62.0</td>
   <td class="num" data-value="67.66233766233766" title="mmlu_prox_de: 70.0 ± 2.7 (n=300)&#10;include_de_gen: 63.0 ± 3.9 (n=154)&#10;">67.7</td>
   <td class="num" data-value="21.360681311630678">21</td>
+  <td class="num" data-value="21.121219019103503" title="overall 62.6 t/s with 3 parallel requests, from 5 of 9 tasks">21</td>
   <td class="num" data-value="5258.027383235864">5258</td>
   <td class="num" data-value="763.404164230451">763</td>
   <td data-value="2026-09-18">2026-09-18</td>
@@ -156,6 +197,7 @@
   <td class="num" data-value="53.0" title="livecodebench_gen: 53.0 ± 3.5 (n=200)&#10;">53.0</td>
   <td class="num" data-value="70.5108225108225" title="mmlu_prox_de: 72.0 ± 2.6 (n=300)&#10;include_de_gen: 67.5 ± 3.8 (n=154)&#10;">70.5</td>
   <td class="num" data-value="83.7975869946743">84</td>
+  <td class="num" data-value="70.42131680616991" title="overall 213.6 t/s with 3 parallel requests, from 5 of 9 tasks">70</td>
   <td class="num" data-value="7236.849600555504">7237</td>
   <td class="num" data-value="554.1085170116276">554</td>
   <td data-value="2026-09-17">2026-09-17</td>
@@ -172,17 +214,51 @@
   <td class="num" data-value="46.5" title="livecodebench_gen: 46.5 ± 3.5 (n=200)&#10;">46.5</td>
   <td class="num" data-value="67.00144300144301" title="mmlu_prox_de: 69.3 ± 2.7 (n=300)&#10;include_de_gen: 62.3 ± 3.9 (n=154)&#10;">67.0</td>
   <td class="num" data-value="31.530246994028012">32</td>
+  <td class="num" data-value="35.229241139471675" title="overall 106.0 t/s with 3 parallel requests, from 5 of 9 tasks">35</td>
   <td class="num" data-value="6290.925546819461">6291</td>
   <td class="num" data-value="639.6515059750527">640</td>
   <td data-value="2026-09-18">2026-09-18</td>
 </tr>
+<tr data-reasoning="off">
+  <td data-value="zgcm-1-7b">
+    <a href="runs/29.html" class="strong">zgcm-1-7b</a>
+    <div class="sub">NVFP4 · Reasoning off · direct · NVIDIA DGX Spark, 128 GB RAM</div>
+  </td>
+  <td class="num strong" data-value="33.36777195514038">33.4</td>
+  <td class="num" data-value="34.68552188552189" title="mmlu_pro: 25.3 ± 2.5 (n=300)&#10;gpqa_diamond_gen: 32.3 ± 3.3 (n=198)&#10;musr_gen: 46.4 ± 3.2 (n=250)&#10;">34.7</td>
+  <td class="num" data-value="40.0" title="omni_math_gen: 40.0 ± 3.1 (n=250)&#10;">40.0</td>
+  <td class="num" data-value="44.33082706766917" title="ifbench: 33.0 ± 2.7 (n=300)&#10;ifeval: 78.7 ± 3.4 (n=150)&#10;">44.3</td>
+  <td class="num" data-value="23.0" title="livecodebench_gen: 23.0 ± 3.0 (n=200)&#10;">23.0</td>
+  <td class="num" data-value="24.82251082251082" title="mmlu_prox_de: 21.0 ± 2.4 (n=300)&#10;include_de_gen: 32.5 ± 3.8 (n=154)&#10;">24.8</td>
+  <td class="num" data-value="33.67868964254804">34</td>
+  <td class="num" data-value="34.0307978654491" title="overall 102.4 t/s with 3 parallel requests, from 9 of 9 tasks">34</td>
+  <td class="num" data-value="3957.900821130251">3958</td>
+  <td class="num" data-value="1022.5117260124534">1023</td>
+  <td data-value="2026-09-23">2026-09-23</td>
+</tr>
 </tbody>
 </table>
 </div>
+<div class="chart-set">
+<script type="application/json" class="chart-data" id="charts-v3-quick">{"categories": [{"key": "overall", "label": "Overall"}, {"key": "knowledge", "label": "Knowledge \u0026 reasoning"}, {"key": "math", "label": "Math"}, {"key": "instructions", "label": "Instruction following"}, {"key": "code", "label": "Code"}, {"key": "german", "label": "German"}], "missing_tps": false, "reasoning_mark": "\ud83d\udcad", "runs": [{"label": "gpt-oss-120b \u00b7 \ud83d\udcad", "run_id": 18, "scores": {"code": 83.0, "german": 77.4, "instructions": 66.1, "knowledge": 68.5, "math": 70.4, "overall": 73.1}, "tps": 19.6, "tps_source": "run"}, {"label": "qwen3.8-flash-next-ple4", "run_id": 15, "scores": {"code": 75.5, "german": 81.1, "instructions": 49.6, "knowledge": 77.9, "math": 73.2, "overall": 71.5}, "tps": 21.6, "tps_source": "run"}, {"label": "gemma-4-12b-it", "run_id": 21, "scores": {"code": 75.0, "german": 74.5, "instructions": 51.3, "knowledge": 67.5, "math": 60.8, "overall": 65.8}, "tps": 13.8, "tps_source": "run"}, {"label": "gemma-4-12b-it-heretic", "run_id": 22, "scores": {"code": 75.0, "german": 76.5, "instructions": 50.0, "knowledge": 68.4, "math": 59.2, "overall": 65.8}, "tps": 13.8, "tps_source": "run"}, {"label": "qwen3.8-27b-mtp", "run_id": 20, "scores": {"code": 65.0, "german": 76.3, "instructions": 49.5, "knowledge": 69.9, "math": 60.4, "overall": 64.2}, "tps": 22.4, "tps_source": "run"}, {"label": "occamy-1.0", "run_id": 27, "scores": {"code": 63.0, "german": 76.9, "instructions": 44.7, "knowledge": 73.1, "math": 62.4, "overall": 64.0}, "tps": 31.6, "tps_source": "run"}, {"label": "qwen3.8-27b-abliterated", "run_id": 23, "scores": {"code": 59.5, "german": 72.1, "instructions": 46.1, "knowledge": 71.5, "math": 58.0, "overall": 61.4}, "tps": 21.9, "tps_source": "run"}, {"label": "ternary-bonsai-2-27b", "run_id": 26, "scores": {"code": 56.5, "german": 74.5, "instructions": 46.3, "knowledge": 69.5, "math": 59.6, "overall": 61.3}, "tps": 19.3, "tps_source": "run"}, {"label": "gemma-4-12b-it-heretic", "run_id": 17, "scores": {"code": 62.0, "german": 67.7, "instructions": 46.8, "knowledge": 63.7, "math": 48.4, "overall": 57.7}, "tps": 21.1, "tps_source": "run"}, {"label": "nemotron-3.5-lightning-dspark", "run_id": 16, "scores": {"code": 53.0, "german": 70.5, "instructions": 46.0, "knowledge": 55.8, "math": 40.4, "overall": 53.1}, "tps": 70.4, "tps_source": "run"}, {"label": "qwen3.5-9b-heretic-v2-mtp", "run_id": 19, "scores": {"code": 46.5, "german": 67.0, "instructions": 44.1, "knowledge": 65.1, "math": 35.6, "overall": 51.7}, "tps": 35.2, "tps_source": "run"}, {"label": "zgcm-1-7b", "run_id": 29, "scores": {"code": 23.0, "german": 24.8, "instructions": 44.3, "knowledge": 34.7, "math": 40.0, "overall": 33.4}, "tps": 34.0, "tps_source": "run"}], "texts": {"overall": "Overall", "overall_title": "Overall score", "points": "points", "score_axis": "Score (0\u2013100)", "source_run": "measured during the run", "source_test": "from the speed test", "tps_axis": "Generation (tokens/s)", "vs": " vs. generation (tokens/s)"}}</script>
+<p class="segmented" role="group" aria-label="Chart category"><button type="button" data-chart-category="overall" class="active" aria-pressed="true">Overall</button><button type="button" data-chart-category="knowledge" class="" aria-pressed="false">Knowledge &amp; reasoning</button><button type="button" data-chart-category="math" class="" aria-pressed="false">Math</button><button type="button" data-chart-category="instructions" class="" aria-pressed="false">Instruction following</button><button type="button" data-chart-category="code" class="" aria-pressed="false">Code</button><button type="button" data-chart-category="german" class="" aria-pressed="false">German</button></p>
+<div class="charts">
+  <figure class="chart-card">
+    <figcaption><span data-chart-title="bar">Overall score</span> <span class="sub">(💭 = with reasoning)</span></figcaption>
+    <div class="chart-box" data-chart-box="bar" style="height: 430px"><canvas data-chart="bar" role="img" aria-label="Score per model variant; the values are also in the table"></canvas></div>
+  </figure>
+  <figure class="chart-card">
+    <figcaption><span data-chart-title="scatter">Overall score vs. generation (tokens/s)</span> <span class="sub">(💭 = with reasoning)</span></figcaption>
+    <div class="chart-box" style="height: 320px"><canvas data-chart="scatter" role="img" aria-label="Score against generation speed in tokens per second; the values are also in the table"></canvas></div>
+    <p class="hint">x-axis: generation per request measured during the run; hollow points only have the speed-test value.</p>
+  </figure>
+</div>
+</div>
 </section>
-<p class="hint">Scores run from 0 to 100; the best value per category is shown in bold. Hovering a category cell lists its tasks with standard error (±) and number of items (n). Speed: generation in tokens/s, prompt processing at 4k context in tokens/s, time to first token at 4k context in ms. Click a column header to sort.</p>
-<footer class="page-foot">As of 2026-09-21 · scores 0–100 · built with <code>bench publish</code></footer>
+<p class="hint">Scores run from 0 to 100; the best value per category is shown in bold. Hovering a category cell lists its tasks with standard error (±) and number of items (n). Speed: generation in tokens/s from the speed test and measured during the run (gen t/s run, per request), prompt processing at 4k context in tokens/s, time to first token at 4k context in ms. Click a column header to sort.</p>
+<footer class="page-foot">As of 2026-09-23 · scores 0–100 · built with <code>bench publish</code></footer>
 </main>
+<script src="chart.umd.min.js"></script>
 <script src="app.js"></script>
 </body>
 </html>

methodology.html

@@ -117,8 +117,9 @@ on hardware and inference software.</p>
 </tbody>
 </table>
 </div>
-<footer class="page-foot">As of 2026-09-21 · scores 0–100 · built with <code>bench publish</code></footer>
+<footer class="page-foot">As of 2026-09-23 · scores 0–100 · built with <code>bench publish</code></footer>
 </main>
+<script src="chart.umd.min.js"></script>
 <script src="app.js"></script>
 </body>
 </html>

runs/15.html

@@ -188,8 +188,9 @@
 </table>
 </div>
 <p class="hint">Measured directly at the inference server.</p>
-<footer class="page-foot">As of 2026-09-21 · scores 0–100 · built with <code>bench publish</code></footer>
+<footer class="page-foot">As of 2026-09-23 · scores 0–100 · built with <code>bench publish</code></footer>
 </main>
+<script src="../chart.umd.min.js"></script>
 <script src="../app.js"></script>
 </body>
 </html>

runs/16.html

@@ -188,8 +188,9 @@
 </table>
 </div>
 <p class="hint">Measured directly at the inference server.</p>
-<footer class="page-foot">As of 2026-09-21 · scores 0–100 · built with <code>bench publish</code></footer>
+<footer class="page-foot">As of 2026-09-23 · scores 0–100 · built with <code>bench publish</code></footer>
 </main>
+<script src="../chart.umd.min.js"></script>
 <script src="../app.js"></script>
 </body>
 </html>

runs/17.html

@@ -188,8 +188,9 @@
 </table>
 </div>
 <p class="hint">Measured directly at the inference server.</p>
-<footer class="page-foot">As of 2026-09-21 · scores 0–100 · built with <code>bench publish</code></footer>
+<footer class="page-foot">As of 2026-09-23 · scores 0–100 · built with <code>bench publish</code></footer>
 </main>
+<script src="../chart.umd.min.js"></script>
 <script src="../app.js"></script>
 </body>
 </html>

runs/18.html

@@ -126,8 +126,9 @@
 </table>
 </div>
 <p class="hint">Measured directly at the inference server.</p>
-<footer class="page-foot">As of 2026-09-21 · scores 0–100 · built with <code>bench publish</code></footer>
+<footer class="page-foot">As of 2026-09-23 · scores 0–100 · built with <code>bench publish</code></footer>
 </main>
+<script src="../chart.umd.min.js"></script>
 <script src="../app.js"></script>
 </body>
 </html>

runs/19.html

@@ -188,8 +188,9 @@
 </table>
 </div>
 <p class="hint">Measured directly at the inference server.</p>
-<footer class="page-foot">As of 2026-09-21 · scores 0–100 · built with <code>bench publish</code></footer>
+<footer class="page-foot">As of 2026-09-23 · scores 0–100 · built with <code>bench publish</code></footer>
 </main>
+<script src="../chart.umd.min.js"></script>
 <script src="../app.js"></script>
 </body>
 </html>

runs/20.html

@@ -188,8 +188,9 @@
 </table>
 </div>
 <p class="hint">Measured directly at the inference server.</p>
-<footer class="page-foot">As of 2026-09-21 · scores 0–100 · built with <code>bench publish</code></footer>
+<footer class="page-foot">As of 2026-09-23 · scores 0–100 · built with <code>bench publish</code></footer>
 </main>
+<script src="../chart.umd.min.js"></script>
 <script src="../app.js"></script>
 </body>
 </html>

runs/21.html

@@ -188,8 +188,9 @@
 </table>
 </div>
 <p class="hint">Measured directly at the inference server.</p>
-<footer class="page-foot">As of 2026-09-21 · scores 0–100 · built with <code>bench publish</code></footer>
+<footer class="page-foot">As of 2026-09-23 · scores 0–100 · built with <code>bench publish</code></footer>
 </main>
+<script src="../chart.umd.min.js"></script>
 <script src="../app.js"></script>
 </body>
 </html>

runs/22.html

@@ -188,8 +188,9 @@
 </table>
 </div>
 <p class="hint">Measured directly at the inference server.</p>
-<footer class="page-foot">As of 2026-09-21 · scores 0–100 · built with <code>bench publish</code></footer>
+<footer class="page-foot">As of 2026-09-23 · scores 0–100 · built with <code>bench publish</code></footer>
 </main>
+<script src="../chart.umd.min.js"></script>
 <script src="../app.js"></script>
 </body>
 </html>

runs/23.html

@@ -188,8 +188,9 @@
 </table>
 </div>
 <p class="hint">Measured directly at the inference server.</p>
-<footer class="page-foot">As of 2026-09-21 · scores 0–100 · built with <code>bench publish</code></footer>
+<footer class="page-foot">As of 2026-09-23 · scores 0–100 · built with <code>bench publish</code></footer>
 </main>
+<script src="../chart.umd.min.js"></script>
 <script src="../app.js"></script>
 </body>
 </html>

runs/26.html

@@ -0,0 +1,196 @@
+<!doctype html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>ternary-bonsai-2-27b – Local LLM Benchmarks</title>
+<link rel="stylesheet" href="../style.css">
+</head>
+<body>
+<header class="top">
+  <a class="brand" href="../index.html">Local LLM Benchmarks</a>
+  <nav>
+    <a href="../index.html">Overview</a>
+    <a href="../methodology.html">Methodology</a>
+  </nav>
+</header>
+<main>
+<p class="sub"><a href="../index.html#v3-quick">← Overview</a></p>
+<h1>ternary-bonsai-2-27b</h1>
+<dl class="meta">
+  <dt>Quantization</dt><dd>FP8</dd>
+  <dt>Reasoning</dt><dd>off</dd>
+  <dt>Profile</dt><dd>quick</dd>
+  <dt>Suite</dt><dd>v2b</dd>
+  <dt>Date</dt><dd>2026-09-21 – 2026-09-23</dd>
+  <dt>Hardware</dt><dd>NVIDIA DGX Spark, 128 GB RAM</dd>
+  <dt>Inference software</dt><dd>vLLM 0.26.0</dd>
+  <dt>Measured</dt><dd>directly at the inference server</dd>
+  <dt>Overall</dt><dd class="strong">61.3</dd>
+</dl>
+
+<h2>Categories</h2>
+<div class="table-wrap">
+<table>
+<thead><tr><th>Category</th><th class="num">Score</th></tr></thead>
+<tbody>
+<tr><td>Knowledge &amp; reasoning</td><td class="num">69.5</td></tr>
+<tr><td>Math</td><td class="num">59.6</td></tr>
+<tr><td>Instruction following</td><td class="num">46.3</td></tr>
+<tr><td>Code</td><td class="num">56.5</td></tr>
+<tr><td>German</td><td class="num">74.5</td></tr>
+</tbody>
+</table>
+</div>
+
+<h2>Tasks</h2>
+<div class="table-wrap">
+<table>
+<thead><tr><th>Category</th><th>Task</th><th class="num">Score</th><th class="num">±</th><th class="num">n</th></tr></thead>
+<tbody>
+<tr>
+  <td>Knowledge &amp; reasoning</td>
+  <td>mmlu_pro</td>
+  <td class="num">77.7</td>
+  <td class="num">± 2.4</td>
+  <td class="num">300</td>
+</tr>
+<tr>
+  <td>Knowledge &amp; reasoning</td>
+  <td>gpqa_diamond_gen</td>
+  <td class="num">65.7</td>
+  <td class="num">± 3.4</td>
+  <td class="num">198</td>
+</tr>
+<tr>
+  <td>Knowledge &amp; reasoning</td>
+  <td>musr_gen</td>
+  <td class="num">65.2</td>
+  <td class="num">± 3.0</td>
+  <td class="num">250</td>
+</tr>
+<tr>
+  <td>Math</td>
+  <td>omni_math_gen</td>
+  <td class="num">59.6</td>
+  <td class="num">± 3.1</td>
+  <td class="num">250</td>
+</tr>
+<tr>
+  <td>Instruction following</td>
+  <td>ifbench</td>
+  <td class="num">34.3</td>
+  <td class="num">± 2.7</td>
+  <td class="num">300</td>
+</tr>
+<tr>
+  <td>Instruction following</td>
+  <td>ifeval</td>
+  <td class="num">82.7</td>
+  <td class="num">± 3.1</td>
+  <td class="num">150</td>
+</tr>
+<tr>
+  <td>Code</td>
+  <td>livecodebench_gen</td>
+  <td class="num">56.5</td>
+  <td class="num">± 3.5</td>
+  <td class="num">200</td>
+</tr>
+<tr>
+  <td>German</td>
+  <td>mmlu_prox_de</td>
+  <td class="num">77.3</td>
+  <td class="num">± 2.4</td>
+  <td class="num">300</td>
+</tr>
+<tr>
+  <td>German</td>
+  <td>include_de_gen</td>
+  <td class="num">68.8</td>
+  <td class="num">± 3.7</td>
+  <td class="num">154</td>
+</tr>
+</tbody>
+</table>
+</div>
+<p class="hint">± is the standard error; the 95% interval spans roughly 1.96 standard errors around the score.</p>
+
+<h2>Speed</h2>
+<div class="table-wrap">
+<table>
+<thead><tr><th>Measure</th><th class="num">Context (target)</th><th class="num">Context (actual)</th><th class="num">Median</th><th class="num">p10</th><th class="num">p90</th><th class="num">n</th></tr></thead>
+<tbody>
+<tr>
+  <td>Generation [t/s]</td>
+  <td class="num">–</td>
+  <td class="num">–</td>
+  <td class="num">28.5</td>
+  <td class="num">28.5</td>
+  <td class="num">28.5</td>
+  <td class="num">5</td>
+</tr>
+<tr>
+  <td>Prompt processing [t/s]</td>
+  <td class="num">512</td>
+  <td class="num">529</td>
+  <td class="num">622.5</td>
+  <td class="num">602.7</td>
+  <td class="num">638.3</td>
+  <td class="num">5</td>
+</tr>
+<tr>
+  <td>Prompt processing [t/s]</td>
+  <td class="num">4096</td>
+  <td class="num">4024</td>
+  <td class="num">898.1</td>
+  <td class="num">896.5</td>
+  <td class="num">899.6</td>
+  <td class="num">5</td>
+</tr>
+<tr>
+  <td>Prompt processing [t/s]</td>
+  <td class="num">16384</td>
+  <td class="num">16001</td>
+  <td class="num">892.4</td>
+  <td class="num">886.4</td>
+  <td class="num">903.4</td>
+  <td class="num">5</td>
+</tr>
+<tr>
+  <td>Time to first token [ms]</td>
+  <td class="num">512</td>
+  <td class="num">529</td>
+  <td class="num">849.8</td>
+  <td class="num">828.8</td>
+  <td class="num">876.9</td>
+  <td class="num">5</td>
+</tr>
+<tr>
+  <td>Time to first token [ms]</td>
+  <td class="num">4096</td>
+  <td class="num">4024</td>
+  <td class="num">4480.4</td>
+  <td class="num">4473.0</td>
+  <td class="num">4488.6</td>
+  <td class="num">5</td>
+</tr>
+<tr>
+  <td>Time to first token [ms]</td>
+  <td class="num">16384</td>
+  <td class="num">16001</td>
+  <td class="num">17931.1</td>
+  <td class="num">17712.7</td>
+  <td class="num">18051.2</td>
+  <td class="num">5</td>
+</tr>
+</tbody>
+</table>
+</div>
+<p class="hint">Measured directly at the inference server.</p>
+<footer class="page-foot">As of 2026-09-23 · scores 0–100 · built with <code>bench publish</code></footer>
+</main>
+<script src="../chart.umd.min.js"></script>
+<script src="../app.js"></script>
+</body>
+</html>

runs/27.html

@@ -0,0 +1,196 @@
+<!doctype html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>occamy-1.0 – Local LLM Benchmarks</title>
+<link rel="stylesheet" href="../style.css">
+</head>
+<body>
+<header class="top">
+  <a class="brand" href="../index.html">Local LLM Benchmarks</a>
+  <nav>
+    <a href="../index.html">Overview</a>
+    <a href="../methodology.html">Methodology</a>
+  </nav>
+</header>
+<main>
+<p class="sub"><a href="../index.html#v3-quick">← Overview</a></p>
+<h1>occamy-1.0</h1>
+<dl class="meta">
+  <dt>Quantization</dt><dd>NVFP4</dd>
+  <dt>Reasoning</dt><dd>off</dd>
+  <dt>Profile</dt><dd>quick</dd>
+  <dt>Suite</dt><dd>v2b</dd>
+  <dt>Date</dt><dd>2026-09-21 – 2026-09-22</dd>
+  <dt>Hardware</dt><dd>NVIDIA DGX Spark, 128 GB RAM</dd>
+  <dt>Inference software</dt><dd>vLLM 0.26.0</dd>
+  <dt>Measured</dt><dd>directly at the inference server</dd>
+  <dt>Overall</dt><dd class="strong">64.0</dd>
+</dl>
+
+<h2>Categories</h2>
+<div class="table-wrap">
+<table>
+<thead><tr><th>Category</th><th class="num">Score</th></tr></thead>
+<tbody>
+<tr><td>Knowledge &amp; reasoning</td><td class="num">73.1</td></tr>
+<tr><td>Math</td><td class="num">62.4</td></tr>
+<tr><td>Instruction following</td><td class="num">44.7</td></tr>
+<tr><td>Code</td><td class="num">63.0</td></tr>
+<tr><td>German</td><td class="num">76.9</td></tr>
+</tbody>
+</table>
+</div>
+
+<h2>Tasks</h2>
+<div class="table-wrap">
+<table>
+<thead><tr><th>Category</th><th>Task</th><th class="num">Score</th><th class="num">±</th><th class="num">n</th></tr></thead>
+<tbody>
+<tr>
+  <td>Knowledge &amp; reasoning</td>
+  <td>mmlu_pro</td>
+  <td class="num">84.0</td>
+  <td class="num">± 2.1</td>
+  <td class="num">300</td>
+</tr>
+<tr>
+  <td>Knowledge &amp; reasoning</td>
+  <td>gpqa_diamond_gen</td>
+  <td class="num">71.7</td>
+  <td class="num">± 3.2</td>
+  <td class="num">198</td>
+</tr>
+<tr>
+  <td>Knowledge &amp; reasoning</td>
+  <td>musr_gen</td>
+  <td class="num">63.6</td>
+  <td class="num">± 3.0</td>
+  <td class="num">250</td>
+</tr>
+<tr>
+  <td>Math</td>
+  <td>omni_math_gen</td>
+  <td class="num">62.4</td>
+  <td class="num">± 3.1</td>
+  <td class="num">250</td>
+</tr>
+<tr>
+  <td>Instruction following</td>
+  <td>ifbench</td>
+  <td class="num">31.3</td>
+  <td class="num">± 2.7</td>
+  <td class="num">300</td>
+</tr>
+<tr>
+  <td>Instruction following</td>
+  <td>ifeval</td>
+  <td class="num">85.3</td>
+  <td class="num">± 2.9</td>
+  <td class="num">150</td>
+</tr>
+<tr>
+  <td>Code</td>
+  <td>livecodebench_gen</td>
+  <td class="num">63.0</td>
+  <td class="num">± 3.4</td>
+  <td class="num">200</td>
+</tr>
+<tr>
+  <td>German</td>
+  <td>mmlu_prox_de</td>
+  <td class="num">79.7</td>
+  <td class="num">± 2.3</td>
+  <td class="num">300</td>
+</tr>
+<tr>
+  <td>German</td>
+  <td>include_de_gen</td>
+  <td class="num">71.4</td>
+  <td class="num">± 3.7</td>
+  <td class="num">154</td>
+</tr>
+</tbody>
+</table>
+</div>
+<p class="hint">± is the standard error; the 95% interval spans roughly 1.96 standard errors around the score.</p>
+
+<h2>Speed</h2>
+<div class="table-wrap">
+<table>
+<thead><tr><th>Measure</th><th class="num">Context (target)</th><th class="num">Context (actual)</th><th class="num">Median</th><th class="num">p10</th><th class="num">p90</th><th class="num">n</th></tr></thead>
+<tbody>
+<tr>
+  <td>Generation [t/s]</td>
+  <td class="num">–</td>
+  <td class="num">–</td>
+  <td class="num">36.7</td>
+  <td class="num">36.7</td>
+  <td class="num">36.7</td>
+  <td class="num">5</td>
+</tr>
+<tr>
+  <td>Prompt processing [t/s]</td>
+  <td class="num">512</td>
+  <td class="num">529</td>
+  <td class="num">3523.5</td>
+  <td class="num">3387.4</td>
+  <td class="num">3545.5</td>
+  <td class="num">5</td>
+</tr>
+<tr>
+  <td>Prompt processing [t/s]</td>
+  <td class="num">4096</td>
+  <td class="num">4024</td>
+  <td class="num">5539.4</td>
+  <td class="num">5491.9</td>
+  <td class="num">5551.9</td>
+  <td class="num">5</td>
+</tr>
+<tr>
+  <td>Prompt processing [t/s]</td>
+  <td class="num">16384</td>
+  <td class="num">16001</td>
+  <td class="num">5467.9</td>
+  <td class="num">5449.6</td>
+  <td class="num">5483.6</td>
+  <td class="num">5</td>
+</tr>
+<tr>
+  <td>Time to first token [ms]</td>
+  <td class="num">512</td>
+  <td class="num">529</td>
+  <td class="num">150.1</td>
+  <td class="num">149.2</td>
+  <td class="num">156.0</td>
+  <td class="num">5</td>
+</tr>
+<tr>
+  <td>Time to first token [ms]</td>
+  <td class="num">4096</td>
+  <td class="num">4024</td>
+  <td class="num">726.4</td>
+  <td class="num">724.8</td>
+  <td class="num">732.7</td>
+  <td class="num">5</td>
+</tr>
+<tr>
+  <td>Time to first token [ms]</td>
+  <td class="num">16384</td>
+  <td class="num">16001</td>
+  <td class="num">2926.3</td>
+  <td class="num">2918.0</td>
+  <td class="num">2936.2</td>
+  <td class="num">5</td>
+</tr>
+</tbody>
+</table>
+</div>
+<p class="hint">Measured directly at the inference server.</p>
+<footer class="page-foot">As of 2026-09-23 · scores 0–100 · built with <code>bench publish</code></footer>
+</main>
+<script src="../chart.umd.min.js"></script>
+<script src="../app.js"></script>
+</body>
+</html>

runs/29.html

@@ -0,0 +1,196 @@
+<!doctype html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>zgcm-1-7b – Local LLM Benchmarks</title>
+<link rel="stylesheet" href="../style.css">
+</head>
+<body>
+<header class="top">
+  <a class="brand" href="../index.html">Local LLM Benchmarks</a>
+  <nav>
+    <a href="../index.html">Overview</a>
+    <a href="../methodology.html">Methodology</a>
+  </nav>
+</header>
+<main>
+<p class="sub"><a href="../index.html#v3-quick">← Overview</a></p>
+<h1>zgcm-1-7b</h1>
+<dl class="meta">
+  <dt>Quantization</dt><dd>NVFP4</dd>
+  <dt>Reasoning</dt><dd>off</dd>
+  <dt>Profile</dt><dd>quick</dd>
+  <dt>Suite</dt><dd>v2b</dd>
+  <dt>Date</dt><dd>2026-09-23</dd>
+  <dt>Hardware</dt><dd>NVIDIA DGX Spark, 128 GB RAM</dd>
+  <dt>Inference software</dt><dd>vLLM 0.26.0</dd>
+  <dt>Measured</dt><dd>directly at the inference server</dd>
+  <dt>Overall</dt><dd class="strong">33.4</dd>
+</dl>
+
+<h2>Categories</h2>
+<div class="table-wrap">
+<table>
+<thead><tr><th>Category</th><th class="num">Score</th></tr></thead>
+<tbody>
+<tr><td>Knowledge &amp; reasoning</td><td class="num">34.7</td></tr>
+<tr><td>Math</td><td class="num">40.0</td></tr>
+<tr><td>Instruction following</td><td class="num">44.3</td></tr>
+<tr><td>Code</td><td class="num">23.0</td></tr>
+<tr><td>German</td><td class="num">24.8</td></tr>
+</tbody>
+</table>
+</div>
+
+<h2>Tasks</h2>
+<div class="table-wrap">
+<table>
+<thead><tr><th>Category</th><th>Task</th><th class="num">Score</th><th class="num">±</th><th class="num">n</th></tr></thead>
+<tbody>
+<tr>
+  <td>Knowledge &amp; reasoning</td>
+  <td>mmlu_pro</td>
+  <td class="num">25.3</td>
+  <td class="num">± 2.5</td>
+  <td class="num">300</td>
+</tr>
+<tr>
+  <td>Knowledge &amp; reasoning</td>
+  <td>gpqa_diamond_gen</td>
+  <td class="num">32.3</td>
+  <td class="num">± 3.3</td>
+  <td class="num">198</td>
+</tr>
+<tr>
+  <td>Knowledge &amp; reasoning</td>
+  <td>musr_gen</td>
+  <td class="num">46.4</td>
+  <td class="num">± 3.2</td>
+  <td class="num">250</td>
+</tr>
+<tr>
+  <td>Math</td>
+  <td>omni_math_gen</td>
+  <td class="num">40.0</td>
+  <td class="num">± 3.1</td>
+  <td class="num">250</td>
+</tr>
+<tr>
+  <td>Instruction following</td>
+  <td>ifbench</td>
+  <td class="num">33.0</td>
+  <td class="num">± 2.7</td>
+  <td class="num">300</td>
+</tr>
+<tr>
+  <td>Instruction following</td>
+  <td>ifeval</td>
+  <td class="num">78.7</td>
+  <td class="num">± 3.4</td>
+  <td class="num">150</td>
+</tr>
+<tr>
+  <td>Code</td>
+  <td>livecodebench_gen</td>
+  <td class="num">23.0</td>
+  <td class="num">± 3.0</td>
+  <td class="num">200</td>
+</tr>
+<tr>
+  <td>German</td>
+  <td>mmlu_prox_de</td>
+  <td class="num">21.0</td>
+  <td class="num">± 2.4</td>
+  <td class="num">300</td>
+</tr>
+<tr>
+  <td>German</td>
+  <td>include_de_gen</td>
+  <td class="num">32.5</td>
+  <td class="num">± 3.8</td>
+  <td class="num">154</td>
+</tr>
+</tbody>
+</table>
+</div>
+<p class="hint">± is the standard error; the 95% interval spans roughly 1.96 standard errors around the score.</p>
+
+<h2>Speed</h2>
+<div class="table-wrap">
+<table>
+<thead><tr><th>Measure</th><th class="num">Context (target)</th><th class="num">Context (actual)</th><th class="num">Median</th><th class="num">p10</th><th class="num">p90</th><th class="num">n</th></tr></thead>
+<tbody>
+<tr>
+  <td>Generation [t/s]</td>
+  <td class="num">–</td>
+  <td class="num">–</td>
+  <td class="num">33.7</td>
+  <td class="num">33.7</td>
+  <td class="num">33.7</td>
+  <td class="num">5</td>
+</tr>
+<tr>
+  <td>Prompt processing [t/s]</td>
+  <td class="num">512</td>
+  <td class="num">526</td>
+  <td class="num">3487.7</td>
+  <td class="num">3466.3</td>
+  <td class="num">3494.1</td>
+  <td class="num">5</td>
+</tr>
+<tr>
+  <td>Prompt processing [t/s]</td>
+  <td class="num">4096</td>
+  <td class="num">4047</td>
+  <td class="num">3957.9</td>
+  <td class="num">3932.1</td>
+  <td class="num">3965.1</td>
+  <td class="num">5</td>
+</tr>
+<tr>
+  <td>Prompt processing [t/s]</td>
+  <td class="num">16384</td>
+  <td class="num">16107</td>
+  <td class="num">3928.9</td>
+  <td class="num">3924.4</td>
+  <td class="num">3936.6</td>
+  <td class="num">5</td>
+</tr>
+<tr>
+  <td>Time to first token [ms]</td>
+  <td class="num">512</td>
+  <td class="num">526</td>
+  <td class="num">150.8</td>
+  <td class="num">150.5</td>
+  <td class="num">151.7</td>
+  <td class="num">5</td>
+</tr>
+<tr>
+  <td>Time to first token [ms]</td>
+  <td class="num">4096</td>
+  <td class="num">4047</td>
+  <td class="num">1022.5</td>
+  <td class="num">1020.7</td>
+  <td class="num">1029.2</td>
+  <td class="num">5</td>
+</tr>
+<tr>
+  <td>Time to first token [ms]</td>
+  <td class="num">16384</td>
+  <td class="num">16107</td>
+  <td class="num">4099.6</td>
+  <td class="num">4091.6</td>
+  <td class="num">4104.3</td>
+  <td class="num">5</td>
+</tr>
+</tbody>
+</table>
+</div>
+<p class="hint">Measured directly at the inference server.</p>
+<footer class="page-foot">As of 2026-09-23 · scores 0–100 · built with <code>bench publish</code></footer>
+</main>
+<script src="../chart.umd.min.js"></script>
+<script src="../app.js"></script>
+</body>
+</html>

style.css

@@ -12,6 +12,7 @@
   --series-1: #2a78d6;
   --series-2: #eb6834;
   --series-3: #1baf7a;
+  --zone: rgba(27, 175, 122, 0.12);
   --warn: #fab219;
   --warn-ink: #7a5200;
   --danger: #c62828;
@@ -38,6 +39,7 @@
     --series-1: #3987e5;
     --series-2: #d95926;
     --series-3: #199e70;
+    --zone: rgba(76, 195, 138, 0.14);
     --warn-ink: #fab219;
     --danger: #ef5350;
     --ok: #4cc38a;
@@ -110,6 +112,9 @@ th[aria-sort="descending"]::after { content: " ▼"; }
 .chart-card { margin: 0 0 16px; padding: 12px; background: var(--surface); border: 1px solid var(--border); border-radius: 8px; }
 figcaption { font-weight: 600; margin-bottom: 8px; }
 .chart-box { position: relative; width: 100%; }
+.thresholds { display: flex; flex-wrap: wrap; gap: 8px 12px; align-items: center; margin: 0; }
+.thresholds label { display: inline-flex; gap: 6px; align-items: center; color: var(--ink-2); }
+.thresholds input { width: 6em; }
 .meta { display: grid; grid-template-columns: max-content 1fr; gap: 4px 16px; padding: 12px; background: var(--surface); border: 1px solid var(--border); border-radius: 8px; }
 .meta dt { color: var(--ink-2); }
 .meta dd { margin: 0; overflow-wrap: anywhere; }