ybrault commited on
Commit
bf9a6ea
·
verified ·
1 Parent(s): 4f827d8

fix(arena): align model presentation, prices and trust links

Browse files
Files changed (1) hide show
  1. index.html +41 -115
index.html CHANGED
@@ -769,6 +769,7 @@
769
  padding: 18px 0 0;
770
  }
771
 
 
772
  .footer-links {
773
  display: flex;
774
  flex-wrap: wrap;
@@ -908,11 +909,11 @@
908
  </p>
909
  <div class="proof-grid" aria-label="Models and modalities">
910
  <div class="proof-chip">
911
- <strong>Qwen3-Coder-Next FP8</strong>
912
- <span>Code and technical generation</span>
913
  </div>
914
  <div class="proof-chip">
915
- <strong>Gemma 4 31B FP8</strong>
916
  <span>Text, vision and extraction</span>
917
  </div>
918
  <div class="proof-chip">
@@ -938,11 +939,11 @@
938
  </div>
939
  <div class="status-metric">
940
  <span>Input</span>
941
- <strong>$0.018 / M in</strong>
942
  </div>
943
  <div class="status-metric">
944
  <span>Output</span>
945
- <strong>$0.16 / M out</strong>
946
  </div>
947
  <a class="status-link" href="#view-defi">Launch Arena</a>
948
  <p class="status-cta-note">100M free tokens</p>
@@ -1056,101 +1057,18 @@
1056
  </section>
1057
 
1058
  <section class="view" id="view-speed" hidden>
1059
- <div class="stat-strip">Reference target: 6x faster TTFT</div>
1060
- <div class="benchmark-grid">
1061
- <article class="panel benchmark-card">
1062
- <p class="section-kicker">SovAPI TTFT</p>
1063
- <strong>0.12 s</strong>
1064
- <p>Target path for code and technical generation.</p>
1065
- </article>
1066
- <article class="panel benchmark-card">
1067
- <p class="section-kicker">Code throughput</p>
1068
- <strong>184 tok/s</strong>
1069
- <p>Reference throughput on the SovAPI Code lane.</p>
1070
- </article>
1071
- <article class="panel benchmark-card">
1072
- <p class="section-kicker">Reasoning throughput</p>
1073
- <strong>234 tok/s</strong>
1074
- <p>Reference throughput on the SovAPI Text lane (analysis, reasoning, structure).</p>
1075
- </article>
1076
- </div>
1077
- <div class="panel" style="margin-top: 16px">
1078
- <h2>Latency reference table</h2>
1079
- <p class="reference-note">
1080
- <strong>Reference / expected values.</strong> The Arena verdict above is recalculated
1081
- from each live duel.
1082
- </p>
1083
- <div class="table-scroll" role="region" aria-label="Latency reference table" tabindex="0">
1084
- <table>
1085
- <thead>
1086
- <tr><th>Scenario</th><th>SovAPI route</th><th>SovAPI reference</th><th>Challenger reference</th><th>Signal</th></tr>
1087
- </thead>
1088
- <tbody>
1089
- <tr><td>TTFT</td><td>All routes</td><td>0.12 &ndash; 0.18 s</td><td>0.74 &ndash; 1.11 s</td><td>~6x faster</td></tr>
1090
- <tr><td>Code throughput</td><td>SovAPI Code</td><td>~184 tok/s</td><td>~84 &ndash; 96 tok/s</td><td>~2x more tokens/s</td></tr>
1091
- <tr><td>Analysis throughput</td><td>SovAPI Text</td><td>~234 tok/s</td><td>~87 &ndash; 100 tok/s</td><td>~2.4x more tokens/s</td></tr>
1092
- <tr><td>Reasoning throughput</td><td>SovAPI Text</td><td>~234 tok/s</td><td>~74 &ndash; 96 tok/s</td><td>~2.5x more tokens/s</td></tr>
1093
- <tr><td>Structure / Vision throughput</td><td>SovAPI Text</td><td>~234 tok/s</td><td>~87 &ndash; 100 tok/s</td><td>~2.4x more tokens/s</td></tr>
1094
- <tr><td>Tool calling / Function call</td><td>SovAPI Code</td><td>~184 tok/s</td><td>~84 &ndash; 96 tok/s</td><td>~2x more tokens/s</td></tr>
1095
- <tr><td>Speech-to-text</td><td>SovAPI Voice</td><td>~391 ms</td><td>~1.0 &ndash; 1.4 s</td><td>~3x faster</td></tr>
1096
- <tr><td>Embeddings + vector search</td><td>SovAPI Memory</td><td>~189 ms</td><td>~350 &ndash; 500 ms</td><td>~2x faster</td></tr>
1097
- <tr><td>End-to-end streaming</td><td>Modeless SovAPI</td><td>SSE direct</td><td>Official provider API</td><td>Measured live only in Arena</td></tr>
1098
- </tbody>
1099
- </table>
1100
- </div>
1101
- </div>
1102
- <div class="panel" style="margin-top: 16px">
1103
- <p>
1104
- <strong>Engine.</strong> Four SovAPI products routed automatically to the model best
1105
- suited to your task: SovAPI Code (code + tool calling), SovAPI Text (analysis,
1106
- reasoning, structure, vision), SovAPI Voice (speech-to-text), SovAPI Memory
1107
- (embeddings + long-term memory). EU sovereign hosting, open-weight Apache 2.0.
1108
- Powered by Qwen3-Coder-Next, Gemma 4 31B, Whisper Large V3, BGE-M3.
1109
- </p>
1110
- </div>
1111
  </section>
1112
-
1113
  <section class="view" id="view-best" hidden>
1114
- <div class="stat-strip">ANTHROPIC-CLASS QUALITY · 60x cheaper than Claude Sonnet 5 · IN EUROPE</div>
1115
- <div class="panel">
1116
- <h2>SovAPI vs Claude Sonnet 5</h2>
1117
- <p class="reference-note">
1118
- <strong>Verified benchmark values.</strong> Fast-inference mode (Non-Think) for
1119
- developer and interactive workloads. Live Arena results remain measured separately
1120
- for each prompt.
1121
- </p>
1122
- <div class="table-scroll" role="region" aria-label="SovAPI versus Sonnet 5 benchmarks" tabindex="0">
1123
- <table>
1124
- <thead>
1125
- <tr><th>Benchmark</th><th>SovAPI</th><th>Sonnet 5</th><th>Gap</th></tr>
1126
- </thead>
1127
- <tbody>
1128
- <tr><td>GPQA Diamond</td><td>87%</td><td>~83%</td><td>+4</td></tr>
1129
- <tr><td>AIME 2025</td><td>91%</td><td>~78%</td><td>+13</td></tr>
1130
- <tr><td>MMLU Pro</td><td>86%</td><td>~86%</td><td>=</td></tr>
1131
- <tr><td>LiveCodeBench Non-Think</td><td>77.1%</td><td>64.8%</td><td>+12.3 pts</td></tr>
1132
- <tr><td>SWE-bench Verified</td><td>~70%</td><td>~72%</td><td>-2</td></tr>
1133
- <tr><td>BFCL v2</td><td>~89%</td><td>~88%</td><td>+1</td></tr>
1134
- </tbody>
1135
- </table>
1136
- </div>
1137
- </div>
1138
- <div class="panel" style="margin-top: 16px">
1139
- <h2>SovAPI vs DeepSeek V4 (Flash &amp; Pro)</h2>
1140
- <p class="reference-note">
1141
- <strong>Verified benchmark values.</strong> Fast-inference mode (Non-Think) for both
1142
- V4-Flash (13B activated) and V4-Pro (49B activated). Gap measured vs V4-Pro.
1143
  </p>
1144
- <div class="table-scroll" role="region" aria-label="SovAPI versus DeepSeek benchmarks" tabindex="0">
1145
- <table>
1146
- <thead>
1147
- <tr><th>Benchmark</th><th>Qwen3</th><th>Gemma 4</th><th>SovAPI</th><th>V4-Flash</th><th>V4-Pro</th><th>Gap</th></tr>
1148
- </thead>
1149
- <tbody>
1150
- <tr><td>LiveCodeBench Non-Think</td><td>71.4%</td><td>80.0%</td><td>77.1%</td><td>55.2%</td><td>56.8%</td><td>+20.3 pts</td></tr>
1151
- </tbody>
1152
- </table>
1153
- </div>
1154
  </div>
1155
  </section>
1156
 
@@ -1158,18 +1076,23 @@
1158
  <div class="stat-strip">Usage-based pricing in US dollars, with no commitment</div>
1159
  <div class="benchmark-grid">
1160
  <article class="panel benchmark-card">
1161
- <p class="section-kicker">Input</p>
1162
- <strong>$0.018</strong>
1163
- <p>Per million input tokens.</p>
 
 
 
 
 
1164
  </article>
1165
  <article class="panel benchmark-card">
1166
- <p class="section-kicker">Output</p>
1167
- <strong>$0.16</strong>
1168
- <p>Per million output tokens.</p>
1169
  </article>
1170
  <article class="panel benchmark-card">
1171
  <p class="section-kicker">Trial grant</p>
1172
- <strong>100 M</strong>
1173
  <p>Trial tokens for developers starting from API ACCESS.</p>
1174
  </article>
1175
  </div>
@@ -1186,11 +1109,12 @@
1186
  <tr><th>Metric</th><th>SovAPI</th><th>Challenger</th><th>How it is shown</th></tr>
1187
  </thead>
1188
  <tbody>
1189
- <tr><td>Input token price</td><td>$0.018 / M reference grid</td><td>-- when unavailable</td><td>No inferred challenger price</td></tr>
1190
- <tr><td>Output token price</td><td>$0.16 / M reference grid</td><td>-- when unavailable</td><td>No inferred challenger price</td></tr>
 
1191
  <tr><td>Per-request cost</td><td>Calculated from real tokens</td><td>-- when unavailable</td><td>Only a server-reported amount is displayed</td></tr>
1192
- <tr><td>Trial grant</td><td>100 M tokens</td><td>None</td><td>API ACCESS onboarding</td></tr>
1193
- <tr><td>Rate limit</td><td>100 req/min free; 600 req/min PAYG</td><td>Provider-specific</td><td>Applied server-side by tier</td></tr>
1194
  <tr><td>Billing</td><td>Exclusively token usage</td><td>Provider-specific</td><td>No SovAPI commitment</td></tr>
1195
  </tbody>
1196
  </table>
@@ -1230,7 +1154,7 @@
1230
  <tbody>
1231
  <tr>
1232
  <td>Free</td>
1233
- <td>100 M tokens</td>
1234
  <td>100 req/min</td>
1235
  <td>Included</td>
1236
  <td>Included</td>
@@ -1241,8 +1165,8 @@
1241
  <td>Pay-as-you-go</td>
1242
  <td>Unlimited</td>
1243
  <td>600 req/min</td>
1244
- <td>$0.018 / M tokens</td>
1245
- <td>$0.16 / M tokens</td>
1246
  <td>Email within 24h</td>
1247
  <td>None</td>
1248
  </tr>
@@ -1291,6 +1215,8 @@ const client = new OpenAI({
1291
  <footer class="footer">
1292
  <span>SOVINFRA - FPLC SAS, 10 Place Vendome, Paris.</span>
1293
  <div class="footer-links">
 
 
1294
  <a href="https://status.sovinfra.cloud/" rel="noopener noreferrer" target="_blank">Status</a>
1295
  <a href="https://codeberg.org/sovinfra" rel="noopener noreferrer" target="_blank">Codeberg</a>
1296
  <a href="/legal">Legal</a>
@@ -1465,10 +1391,10 @@ const client = new OpenAI({
1465
  reasoning: "Reasoning",
1466
  });
1467
  const SOVAPI_MODEL_DISPLAY_NAMES = Object.freeze({
1468
- "qwen3-coder-next": "Qwen3-Coder-Next",
1469
- "qwen3-coder-next-fp8": "Qwen3-Coder-Next",
1470
- "gemma-4-31b": "Gemma 4 31B FP8",
1471
- "gemma-4-31b-it-fp8": "Gemma 4 31B FP8",
1472
  });
1473
  const state = {
1474
  sovapi: null,
 
769
  padding: 18px 0 0;
770
  }
771
 
772
+ .benchmark-sources { display: flex; flex-wrap: wrap; gap: 14px 24px; }
773
  .footer-links {
774
  display: flex;
775
  flex-wrap: wrap;
 
909
  </p>
910
  <div class="proof-grid" aria-label="Models and modalities">
911
  <div class="proof-chip">
912
+ <strong>Qwen3.8-27B FP8</strong>
913
+ <span>Code, reasoning, vision and tool calling</span>
914
  </div>
915
  <div class="proof-chip">
916
+ <strong>Gemma 4 31B BF16 · H200</strong>
917
  <span>Text, vision and extraction</span>
918
  </div>
919
  <div class="proof-chip">
 
939
  </div>
940
  <div class="status-metric">
941
  <span>Input</span>
942
+ <strong>$0.12 / M in</strong>
943
  </div>
944
  <div class="status-metric">
945
  <span>Output</span>
946
+ <strong>$0.38 / M out</strong>
947
  </div>
948
  <a class="status-link" href="#view-defi">Launch Arena</a>
949
  <p class="status-cta-note">100M free tokens</p>
 
1057
  </section>
1058
 
1059
  <section class="view" id="view-speed" hidden>
1060
+ <div class="panel"><h2>Measure latency in the Arena</h2><p>Run the same prompt to compare time to first token, throughput and request cost. Results are measured for each request.</p><button class="nav-link" type="button" data-nav-tab="defi">Run the Arena →</button></div>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1061
  </section>
 
1062
  <section class="view" id="view-best" hidden>
1063
+ <div class="panel"><h2>Model developer benchmarks</h2>
1064
+ <p>Results published by the model developers, according to their evaluation protocols.</p>
1065
+ <p class="benchmark-sources">
1066
+ <a href="https://qwen.ai/research">Qwen research</a>
1067
+ <a href="https://deepmind.google/models/gemma/">Google Gemma</a>
1068
+ <a href="https://www.anthropic.com/news">Anthropic publications</a>
1069
+ <a href="https://github.com/deepseek-ai">DeepSeek publications</a>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1070
  </p>
1071
+ <p>Compare your own workloads in the Arena.</p>
 
 
 
 
 
 
 
 
 
1072
  </div>
1073
  </section>
1074
 
 
1076
  <div class="stat-strip">Usage-based pricing in US dollars, with no commitment</div>
1077
  <div class="benchmark-grid">
1078
  <article class="panel benchmark-card">
1079
+ <p class="section-kicker">Gemma 4 31B</p>
1080
+ <strong>$0.12 · $0.38</strong>
1081
+ <p>Per million input and output tokens. No cache-read tariff.</p>
1082
+ </article>
1083
+ <article class="panel benchmark-card">
1084
+ <p class="section-kicker">Qwen3.8-27B</p>
1085
+ <strong>$0.12 · $0.38</strong>
1086
+ <p>Per million input and output tokens.</p>
1087
  </article>
1088
  <article class="panel benchmark-card">
1089
+ <p class="section-kicker">Qwen cache read</p>
1090
+ <strong>$0.04 cached</strong>
1091
+ <p>Per million cached input tokens.</p>
1092
  </article>
1093
  <article class="panel benchmark-card">
1094
  <p class="section-kicker">Trial grant</p>
1095
+ <strong>1 B</strong>
1096
  <p>Trial tokens for developers starting from API ACCESS.</p>
1097
  </article>
1098
  </div>
 
1109
  <tr><th>Metric</th><th>SovAPI</th><th>Challenger</th><th>How it is shown</th></tr>
1110
  </thead>
1111
  <tbody>
1112
+ <tr><td>Gemma 4 31B</td><td>$0.12 / M input · $0.38 / M output</td><td>-- when unavailable</td><td>No cache-read tariff</td></tr>
1113
+ <tr><td>Qwen3.8-27B</td><td>$0.12 / M input · $0.38 / M output</td><td>-- when unavailable</td><td>Model-specific live cost</td></tr>
1114
+ <tr><td>Qwen cache read</td><td>$0.04 cached</td><td>-- when unavailable</td><td>Never inferred from uncached input</td></tr>
1115
  <tr><td>Per-request cost</td><td>Calculated from real tokens</td><td>-- when unavailable</td><td>Only a server-reported amount is displayed</td></tr>
1116
+ <tr><td>Trial grant</td><td>1 B tokens · Qwen3.8-27B and Gemma 4 31B</td><td>None</td><td>API ACCESS onboarding</td></tr>
1117
+ <tr><td>Rate limit</td><td>100 req/min free; 600 req/min PAYG</td><td>Provider-specific</td><td>Applied server-side per key</td></tr>
1118
  <tr><td>Billing</td><td>Exclusively token usage</td><td>Provider-specific</td><td>No SovAPI commitment</td></tr>
1119
  </tbody>
1120
  </table>
 
1154
  <tbody>
1155
  <tr>
1156
  <td>Free</td>
1157
+ <td>1 B tokens · Qwen3.8-27B and Gemma 4 31B</td>
1158
  <td>100 req/min</td>
1159
  <td>Included</td>
1160
  <td>Included</td>
 
1165
  <td>Pay-as-you-go</td>
1166
  <td>Unlimited</td>
1167
  <td>600 req/min</td>
1168
+ <td>$0.12 / M tokens · Qwen cached input: $0.04 / M</td>
1169
+ <td>$0.38 / M tokens</td>
1170
  <td>Email within 24h</td>
1171
  <td>None</td>
1172
  </tr>
 
1215
  <footer class="footer">
1216
  <span>SOVINFRA - FPLC SAS, 10 Place Vendome, Paris.</span>
1217
  <div class="footer-links">
1218
+ <a href="https://sovinfra.ai/sovereignty-security">Sovereignty and Security</a>
1219
+ <a href="https://sovinfra.ai/availability-continuity">Availability and Continuity</a>
1220
  <a href="https://status.sovinfra.cloud/" rel="noopener noreferrer" target="_blank">Status</a>
1221
  <a href="https://codeberg.org/sovinfra" rel="noopener noreferrer" target="_blank">Codeberg</a>
1222
  <a href="/legal">Legal</a>
 
1391
  reasoning: "Reasoning",
1392
  });
1393
  const SOVAPI_MODEL_DISPLAY_NAMES = Object.freeze({
1394
+ "qwen3-coder-next": "Qwen3.8-27B",
1395
+ "qwen3-coder-next-fp8": "Qwen3.8-27B",
1396
+ "gemma-4-31b": "Gemma 4 31B BF16 · H200",
1397
+ "gemma-4-31b-it-fp8": "Gemma 4 31B BF16 · H200",
1398
  });
1399
  const state = {
1400
  sovapi: null,