benchmark_comparison.svg

<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 1400 640" width="100%" style="background-color:#f1f5f9; font-family: system-ui, -apple-system, 'Segoe UI', Roboto, Helvetica, Arial, sans-serif;">

  <defs>
    <filter id="shadow" x="-5%" y="-5%" width="115%" height="115%" filterUnits="userSpaceOnUse">
      <feDropShadow dx="0" dy="4" stdDeviation="6" flood-color="#0f172a" flood-opacity="0.05"/>
    </filter>
    <linearGradient id="red-grad" x1="0" y1="0" x2="0" y2="1">
      <stop offset="0%" stop-color="#dc2626"/><stop offset="100%" stop-color="#991b1b"/>
    </linearGradient>
    <linearGradient id="dark-grad" x1="0" y1="0" x2="0" y2="1">
      <stop offset="0%" stop-color="#334155"/><stop offset="100%" stop-color="#0f172a"/>
    </linearGradient>
    <linearGradient id="orange-grad" x1="0" y1="0" x2="0" y2="1">
      <stop offset="0%" stop-color="#ea580c"/><stop offset="100%" stop-color="#9a3412"/>
    </linearGradient>
    <linearGradient id="blue-grad" x1="0" y1="0" x2="0" y2="1">
      <stop offset="0%" stop-color="#0369a1"/><stop offset="100%" stop-color="#075985"/>
    </linearGradient>
    <linearGradient id="green-grad" x1="0" y1="0" x2="0" y2="1">
      <stop offset="0%" stop-color="#059669"/><stop offset="100%" stop-color="#047857"/>
    </linearGradient>
    <linearGradient id="purple-grad" x1="0" y1="0" x2="0" y2="1">
      <stop offset="0%" stop-color="#9333ea"/><stop offset="100%" stop-color="#6b21a8"/>
    </linearGradient>
  </defs>

  <!-- Page Background -->
  <rect width="1400" height="640" fill="#f1f5f9" rx="0"/>

  <!-- ── HEADER ── -->
  <g transform="translate(48,36)">
    <text x="0" y="0" font-size="26" font-weight="800" fill="#0f172a" letter-spacing="-0.5">OpenGCM-v2:
      <tspan font-weight="400" fill="#475569">Performance Evaluation vs. Top-Tier Models</tspan>
    </text>
    <text x="0" y="22" font-size="12" font-weight="500" fill="#94a3b8">June 2026 · N/A = no publicly available score · † tau2-bench telecom community eval</text>
  </g>

  <!-- ── MAIN CARD ── -->
  <rect x="40" y="76" width="1320" height="534" rx="12" fill="#ffffff" filter="url(#shadow)"/>

  <!-- ── LEGEND (inside card, top) ── -->
  <g transform="translate(120, 96)">
    <rect x="0"   y="0" width="16" height="10" rx="2" fill="url(#red-grad)"/>
    <text x="21"  y="9" font-size="11" font-weight="700" fill="#0f172a">OpenGCM-v2 (Ours)</text>

    <rect x="170" y="0" width="16" height="10" rx="2" fill="url(#dark-grad)"/>
    <text x="191" y="9" font-size="11" font-weight="700" fill="#0f172a">GPT-5</text>

    <rect x="255" y="0" width="16" height="10" rx="2" fill="url(#orange-grad)"/>
    <text x="276" y="9" font-size="11" font-weight="700" fill="#0f172a">Claude Opus 4.6</text>

    <rect x="415" y="0" width="16" height="10" rx="2" fill="url(#blue-grad)"/>
    <text x="436" y="9" font-size="11" font-weight="700" fill="#0f172a">GLM-5V-Turbo</text>

    <rect x="570" y="0" width="16" height="10" rx="2" fill="url(#green-grad)"/>
    <text x="591" y="9" font-size="11" font-weight="700" fill="#0f172a">GLM-4.7</text>

    <rect x="670" y="0" width="16" height="10" rx="2" fill="url(#purple-grad)"/>
    <text x="691" y="9" font-size="11" font-weight="700" fill="#0f172a">gemma-4-12B-fable5-v2-tau2</text>
  </g>

  <!--
    CHART COORDINATE SYSTEM
    translate(120, 112) = chart group origin
    Baseline y = 450 (0%)
    Chart top  y = 50  (100%)
    Height     = 400px → 1% = 4.0px
    bar height = value * 4.0
    bar y      = 450 - (value * 4.0)

    6 groups, 6 bars each: bar-w=18, gap=5 → group-w=18*6+5*5=133px
    Group x offsets (translate within chart group):
      AIME-26:        tx=30
      SWE-Bench-Pro:  tx=235
      LiveCodeBench:  tx=440
      HMMT Feb 26:    tx=645
      GPQA Diamond:   tx=850
      Terminal-Bench: tx=1055
  -->

  <g transform="translate(120, 112)">

    <!-- ── Y-AXIS GRIDLINES & LABELS ── -->
    <!-- 100% → y=50 -->
    <line x1="0" y1="50"  x2="1230" y2="50"  stroke="#e2e8f0" stroke-width="1"/>
    <text x="-10" y="54"  font-size="12" font-weight="600" fill="#64748b" text-anchor="end">100</text>
    <!-- 80% → y=130 -->
    <line x1="0" y1="130" x2="1230" y2="130" stroke="#e2e8f0" stroke-dasharray="4 4" stroke-width="1"/>
    <text x="-10" y="134" font-size="12" font-weight="600" fill="#64748b" text-anchor="end">80</text>
    <!-- 60% → y=210 -->
    <line x1="0" y1="210" x2="1230" y2="210" stroke="#e2e8f0" stroke-dasharray="4 4" stroke-width="1"/>
    <text x="-10" y="214" font-size="12" font-weight="600" fill="#64748b" text-anchor="end">60</text>
    <!-- 40% → y=290 -->
    <line x1="0" y1="290" x2="1230" y2="290" stroke="#e2e8f0" stroke-dasharray="4 4" stroke-width="1"/>
    <text x="-10" y="294" font-size="12" font-weight="600" fill="#64748b" text-anchor="end">40</text>
    <!-- 20% → y=370 -->
    <line x1="0" y1="370" x2="1230" y2="370" stroke="#e2e8f0" stroke-dasharray="4 4" stroke-width="1"/>
    <text x="-10" y="374" font-size="12" font-weight="600" fill="#64748b" text-anchor="end">20</text>
    <!-- 0% → y=450 baseline -->
    <line x1="0" y1="450" x2="1230" y2="450" stroke="#475569" stroke-width="2"/>
    <text x="-10" y="454" font-size="12" font-weight="600" fill="#64748b" text-anchor="end">0</text>

    <!-- Y-axis title -->
    <text transform="rotate(-90)" x="-250" y="-58" font-size="13" font-weight="700" fill="#334155" text-anchor="middle">Accuracy / Pass@1 (%)</text>

    <!-- ══════════════════════════════════════════════════════ -->
    <!-- GROUP 1: AIME-26                                       -->
    <!-- OpenGCM 93.2 | GPT-5 92.0 | Claude 88.0              -->
    <!-- GLM-5V N/A   | GLM-4.7 87 | gemma N/A                -->
    <!-- ══════════════════════════════════════════════════════ -->
    <g transform="translate(30,0)">
      <!-- OpenGCM-v2: 93.2 → h=372.8, y=77.2 -->
      <rect x="0"   y="77"  width="18" height="373" rx="3" fill="url(#red-grad)"/>
      <text x="9"   y="69"  font-size="11" font-weight="800" fill="#991b1b" text-anchor="middle">93.2</text>
      <!-- GPT-5: 92.0 → h=368, y=82 -->
      <rect x="23"  y="82"  width="18" height="368" rx="3" fill="url(#dark-grad)"/>
      <text x="32"  y="74"  font-size="10" font-weight="600" fill="#334155" text-anchor="middle">92.0</text>
      <!-- Claude 4.6: 88.0 → h=352, y=98 -->
      <rect x="46"  y="98"  width="18" height="352" rx="3" fill="url(#orange-grad)"/>
      <text x="55"  y="90"  font-size="10" font-weight="600" fill="#9a3412" text-anchor="middle">88.0</text>
      <!-- GLM-5V-Turbo: N/A -->
      <rect x="69"  y="443" width="18" height="7"   rx="3" fill="url(#blue-grad)" opacity="0.35"/>
      <text x="78"  y="438" font-size="9"  fill="#0369a1" text-anchor="middle">N/A</text>
      <!-- GLM-4.7: 87.0 → h=348, y=102 -->
      <rect x="92"  y="102" width="18" height="348" rx="3" fill="url(#green-grad)"/>
      <text x="101" y="94"  font-size="10" font-weight="600" fill="#047857" text-anchor="middle">87.0</text>
      <!-- gemma: N/A -->
      <rect x="115" y="443" width="18" height="7"   rx="3" fill="url(#purple-grad)" opacity="0.35"/>
      <text x="124" y="438" font-size="9"  fill="#7e22ce" text-anchor="middle">N/A</text>

      <!-- Label -->
      <text x="66" y="472" font-size="13" font-weight="700" fill="#0f172a" text-anchor="middle">AIME-26</text>
      <text x="66" y="487" font-size="11" font-weight="500" fill="#64748b" text-anchor="middle">Math Olympiad</text>
      <text x="66" y="500" font-size="10" fill="#94a3b8" text-anchor="middle">(Pass@1)</text>
    </g>

    <!-- ══════════════════════════════════════════════════════ -->
    <!-- GROUP 2: SWE-Bench-Pro                                 -->
    <!-- OpenGCM 35.6 | GPT-5 74.9 | Claude 57.3              -->
    <!-- GLM-5V N/A   | GLM-4.7 38 | gemma N/A                -->
    <!-- ══════════════════════════════════════════════════════ -->
    <g transform="translate(235,0)">
      <!-- OpenGCM-v2: 35.6 → h=142.4, y=307.6 -->
      <rect x="0"   y="308" width="18" height="142" rx="3" fill="url(#red-grad)"/>
      <text x="9"   y="300" font-size="11" font-weight="800" fill="#991b1b" text-anchor="middle">35.6</text>
      <!-- GPT-5: 74.9 → h=299.6, y=150.4 -->
      <rect x="23"  y="150" width="18" height="300" rx="3" fill="url(#dark-grad)"/>
      <text x="32"  y="142" font-size="10" font-weight="600" fill="#334155" text-anchor="middle">74.9</text>
      <!-- Claude 4.6: 57.3 → h=229.2, y=220.8 -->
      <rect x="46"  y="221" width="18" height="229" rx="3" fill="url(#orange-grad)"/>
      <text x="55"  y="213" font-size="10" font-weight="600" fill="#9a3412" text-anchor="middle">57.3</text>
      <!-- GLM-5V: N/A -->
      <rect x="69"  y="443" width="18" height="7"   rx="3" fill="url(#blue-grad)" opacity="0.35"/>
      <text x="78"  y="438" font-size="9"  fill="#0369a1" text-anchor="middle">N/A</text>
      <!-- GLM-4.7: 38.0 → h=152, y=298 -->
      <rect x="92"  y="298" width="18" height="152" rx="3" fill="url(#green-grad)"/>
      <text x="101" y="290" font-size="10" font-weight="600" fill="#047857" text-anchor="middle">38.0</text>
      <!-- gemma: N/A -->
      <rect x="115" y="443" width="18" height="7"   rx="3" fill="url(#purple-grad)" opacity="0.35"/>
      <text x="124" y="438" font-size="9"  fill="#7e22ce" text-anchor="middle">N/A</text>

      <text x="66" y="472" font-size="13" font-weight="700" fill="#0f172a" text-anchor="middle">SWE-Bench</text>
      <text x="66" y="487" font-size="11" font-weight="500" fill="#64748b" text-anchor="middle">Pro</text>
      <text x="66" y="500" font-size="10" fill="#94a3b8" text-anchor="middle">(Resolved %)</text>
    </g>

    <!-- ══════════════════════════════════════════════════════ -->
    <!-- GROUP 3: LiveCodeBench                                  -->
    <!-- OpenGCM 73.1 | GPT-5 86.0 | Claude 71.0              -->
    <!-- GLM-5V N/A   | GLM-4.7 84.9| gemma N/A               -->
    <!-- ══════════════════════════════════════════════════════ -->
    <g transform="translate(440,0)">
      <!-- OpenGCM-v2: 73.1 → h=292.4, y=157.6 -->
      <rect x="0"   y="158" width="18" height="292" rx="3" fill="url(#red-grad)"/>
      <text x="9"   y="150" font-size="11" font-weight="800" fill="#991b1b" text-anchor="middle">73.1</text>
      <!-- GPT-5: 86.0 → h=344, y=106 -->
      <rect x="23"  y="106" width="18" height="344" rx="3" fill="url(#dark-grad)"/>
      <text x="32"  y="98"  font-size="10" font-weight="600" fill="#334155" text-anchor="middle">86.0</text>
      <!-- Claude 4.6: 71.0 → h=284, y=166 -->
      <rect x="46"  y="166" width="18" height="284" rx="3" fill="url(#orange-grad)"/>
      <text x="55"  y="158" font-size="10" font-weight="600" fill="#9a3412" text-anchor="middle">71.0</text>
      <!-- GLM-5V: N/A -->
      <rect x="69"  y="443" width="18" height="7"   rx="3" fill="url(#blue-grad)" opacity="0.35"/>
      <text x="78"  y="438" font-size="9"  fill="#0369a1" text-anchor="middle">N/A</text>
      <!-- GLM-4.7: 84.9 → h=339.6, y=110.4 -->
      <rect x="92"  y="110" width="18" height="340" rx="3" fill="url(#green-grad)"/>
      <text x="101" y="102" font-size="10" font-weight="600" fill="#047857" text-anchor="middle">84.9</text>
      <!-- gemma: N/A -->
      <rect x="115" y="443" width="18" height="7"   rx="3" fill="url(#purple-grad)" opacity="0.35"/>
      <text x="124" y="438" font-size="9"  fill="#7e22ce" text-anchor="middle">N/A</text>

      <text x="66" y="472" font-size="13" font-weight="700" fill="#0f172a" text-anchor="middle">LiveCode</text>
      <text x="66" y="487" font-size="11" font-weight="500" fill="#64748b" text-anchor="middle">Bench</text>
      <text x="66" y="500" font-size="10" fill="#94a3b8" text-anchor="middle">(Pass@1)</text>
    </g>

    <!-- ══════════════════════════════════════════════════════ -->
    <!-- GROUP 4: HMMT Feb 26                                   -->
    <!-- OpenGCM 62.3 | GPT-5 72.0 | Claude 74.3              -->
    <!-- GLM-5V N/A   | GLM-4.7 55 | gemma N/A                -->
    <!-- ══════════════════════════════════════════════════════ -->
    <g transform="translate(645,0)">
      <!-- OpenGCM-v2: 62.3 → h=249.2, y=200.8 -->
      <rect x="0"   y="201" width="18" height="249" rx="3" fill="url(#red-grad)"/>
      <text x="9"   y="193" font-size="11" font-weight="800" fill="#991b1b" text-anchor="middle">62.3</text>
      <!-- GPT-5: 72.0 → h=288, y=162 -->
      <rect x="23"  y="162" width="18" height="288" rx="3" fill="url(#dark-grad)"/>
      <text x="32"  y="154" font-size="10" font-weight="600" fill="#334155" text-anchor="middle">72.0</text>
      <!-- Claude 4.6: 74.3 → h=297.2, y=152.8 -->
      <rect x="46"  y="153" width="18" height="297" rx="3" fill="url(#orange-grad)"/>
      <text x="55"  y="145" font-size="10" font-weight="600" fill="#9a3412" text-anchor="middle">74.3</text>
      <!-- GLM-5V: N/A -->
      <rect x="69"  y="443" width="18" height="7"   rx="3" fill="url(#blue-grad)" opacity="0.35"/>
      <text x="78"  y="438" font-size="9"  fill="#0369a1" text-anchor="middle">N/A</text>
      <!-- GLM-4.7: 55.0 → h=220, y=230 -->
      <rect x="92"  y="230" width="18" height="220" rx="3" fill="url(#green-grad)"/>
      <text x="101" y="222" font-size="10" font-weight="600" fill="#047857" text-anchor="middle">55.0</text>
      <!-- gemma: N/A -->
      <rect x="115" y="443" width="18" height="7"   rx="3" fill="url(#purple-grad)" opacity="0.35"/>
      <text x="124" y="438" font-size="9"  fill="#7e22ce" text-anchor="middle">N/A</text>

      <text x="66" y="472" font-size="13" font-weight="700" fill="#0f172a" text-anchor="middle">HMMT Feb 26</text>
      <text x="66" y="487" font-size="11" font-weight="500" fill="#64748b" text-anchor="middle">Harvard-MIT Math</text>
      <text x="66" y="500" font-size="10" fill="#94a3b8" text-anchor="middle">(Accuracy)</text>
    </g>

    <!-- ══════════════════════════════════════════════════════ -->
    <!-- GROUP 5: GPQA Diamond                                  -->
    <!-- OpenGCM 81.9 | GPT-5 88.4 | Claude 91.3              -->
    <!-- GLM-5V 80.9  | GLM-4.7 85.7| gemma N/A               -->
    <!-- ══════════════════════════════════════════════════════ -->
    <g transform="translate(850,0)">
      <!-- OpenGCM-v2: 81.9 → h=327.6, y=122.4 -->
      <rect x="0"   y="122" width="18" height="328" rx="3" fill="url(#red-grad)"/>
      <text x="9"   y="114" font-size="11" font-weight="800" fill="#991b1b" text-anchor="middle">81.9</text>
      <!-- GPT-5: 88.4 → h=353.6, y=96.4 -->
      <rect x="23"  y="96"  width="18" height="354" rx="3" fill="url(#dark-grad)"/>
      <text x="32"  y="88"  font-size="10" font-weight="600" fill="#334155" text-anchor="middle">88.4</text>
      <!-- Claude 4.6: 91.3 → h=365.2, y=84.8 -->
      <rect x="46"  y="85"  width="18" height="365" rx="3" fill="url(#orange-grad)"/>
      <text x="55"  y="77"  font-size="10" font-weight="600" fill="#9a3412" text-anchor="middle">91.3</text>
      <!-- GLM-5V-Turbo: 80.9 → h=323.6, y=126.4 -->
      <rect x="69"  y="126" width="18" height="324" rx="3" fill="url(#blue-grad)"/>
      <text x="78"  y="118" font-size="10" font-weight="600" fill="#075985" text-anchor="middle">80.9</text>
      <!-- GLM-4.7: 85.7 → h=342.8, y=107.2 -->
      <rect x="92"  y="107" width="18" height="343" rx="3" fill="url(#green-grad)"/>
      <text x="101" y="99"  font-size="10" font-weight="600" fill="#047857" text-anchor="middle">85.7</text>
      <!-- gemma: N/A -->
      <rect x="115" y="443" width="18" height="7"   rx="3" fill="url(#purple-grad)" opacity="0.35"/>
      <text x="124" y="438" font-size="9"  fill="#7e22ce" text-anchor="middle">N/A</text>

      <text x="66" y="472" font-size="13" font-weight="700" fill="#0f172a" text-anchor="middle">GPQA Diamond</text>
      <text x="66" y="487" font-size="11" font-weight="500" fill="#64748b" text-anchor="middle">PhD-Level Science</text>
      <text x="66" y="500" font-size="10" fill="#94a3b8" text-anchor="middle">(Pass@1)</text>
    </g>

    <!-- ══════════════════════════════════════════════════════ -->
    <!-- GROUP 6: Terminal-Bench 2.0                            -->
    <!-- OpenGCM 40.3 | GPT-5 78.0 | Claude 59.3              -->
    <!-- GLM-5V N/A   | GLM-4.7 56.2| gemma 55†              -->
    <!-- ══════════════════════════════════════════════════════ -->
    <g transform="translate(1055,0)">
      <!-- OpenGCM-v2: 40.3 → h=161.2, y=288.8 -->
      <rect x="0"   y="289" width="18" height="161" rx="3" fill="url(#red-grad)"/>
      <text x="9"   y="281" font-size="11" font-weight="800" fill="#991b1b" text-anchor="middle">40.3</text>
      <!-- GPT-5: 78.0 → h=312, y=138 -->
      <rect x="23"  y="138" width="18" height="312" rx="3" fill="url(#dark-grad)"/>
      <text x="32"  y="130" font-size="10" font-weight="600" fill="#334155" text-anchor="middle">78.0</text>
      <!-- Claude 4.6: 59.3 → h=237.2, y=212.8 -->
      <rect x="46"  y="213" width="18" height="237" rx="3" fill="url(#orange-grad)"/>
      <text x="55"  y="205" font-size="10" font-weight="600" fill="#9a3412" text-anchor="middle">59.3</text>
      <!-- GLM-5V: N/A -->
      <rect x="69"  y="443" width="18" height="7"   rx="3" fill="url(#blue-grad)" opacity="0.35"/>
      <text x="78"  y="438" font-size="9"  fill="#0369a1" text-anchor="middle">N/A</text>
      <!-- GLM-4.7: 56.2 → h=224.8, y=225.2 -->
      <rect x="92"  y="225" width="18" height="225" rx="3" fill="url(#green-grad)"/>
      <text x="101" y="217" font-size="10" font-weight="600" fill="#047857" text-anchor="middle">56.2</text>
      <!-- gemma-4-12B: ~55 → h=220, y=230 -->
      <rect x="115" y="230" width="18" height="220" rx="3" fill="url(#purple-grad)" opacity="0.88"/>
      <text x="124" y="222" font-size="10" font-weight="600" fill="#6b21a8" text-anchor="middle">55†</text>

      <text x="66" y="472" font-size="13" font-weight="700" fill="#0f172a" text-anchor="middle">Terminal-Bench</text>
      <text x="66" y="487" font-size="11" font-weight="500" fill="#64748b" text-anchor="middle">2.0</text>
      <text x="66" y="500" font-size="10" fill="#94a3b8" text-anchor="middle">(Accuracy)</text>
    </g>

    <!-- ── CATEGORY BRACKETS ── -->
    <path d="M 20 525 L 20 534 L 495 534 L 495 525" fill="none" stroke="#cbd5e1" stroke-width="1.8"/>
    <text x="257" y="550" font-size="13" font-weight="800" fill="#64748b" text-anchor="middle" letter-spacing="1">KNOWLEDGE &amp; REASONING</text>

    <path d="M 520 525 L 520 534 L 1205 534 L 1205 525" fill="none" stroke="#cbd5e1" stroke-width="1.8"/>
    <text x="862" y="550" font-size="13" font-weight="800" fill="#64748b" text-anchor="middle" letter-spacing="1">CODING &amp; AGENTIC CAPABILITIES</text>

  </g>

  <!-- ── FOOTER ── -->
  <text x="50" y="624" font-size="10" fill="#94a3b8">† gemma-4-12B Terminal score from tau2-bench telecom (community eval by yuxinlu1). Sources: OpenAI, Artificial Analysis, Z.ai Docs, WhatLLM.org — June 2026.</text>

</svg>