Update index.html

index.html

@@ -1120,7 +1120,7 @@ body.dark canvas{filter:none;}
   </p></div>
   <div class="fni"><b>💰 Pricing</b><p>Input cost in $/million tokens. 0 = free open-weights. GPT-5-Nano $0.05/M (cheapest frontier). Qwen3.5-35B $0.10/M = Gemini 2.5 FL-Lite $0.10/M. DeepSeek V3.2 $0.14/M. GPT-5.2 $1.75/M · Claude Opus 4.6 $5/M.</p></div>
   <div class="fni" style="border-left:4px solid #e11d48"><b>📋 Changelog v2.2</b><p>
-  <b>v2.2.3</b> 🧬 <b>Darwin-35B-A3B-Opus</b> (VIDRAFT, 35B MoE 3B active, 2026.03) added — GPQA Diamond 90.0% (surpasses both parents), MMMLU 85.0%, Darwin V5 evolutionary merge + Model MRI. Apache 2.0. 🇰🇷 Korean Sovereign AI. 72 models total.<br>
+  <b>v2.2.3</b> 🧬 <b>Darwin-35B-A3B-Opus</b> (VIDRAFT, 35B MoE 3B active, 2026.03) added — GPQA Diamond 90.0% (surpasses both parents), MMMLU 85.0%, MMLU-Pro 85.0, BFCL 67.3, IFEval 91.9, LCB 74.6 (Father official inferred). Darwin V5 evolutionary merge + Model MRI. 8/10 coverage. Apache 2.0. 72 models total.<br>
   <b>v2.2.2</b> 🆕 <b>NVIDIA Nemotron 3 Super</b> (120B/12B LatentMoE, 2026.03.11) added — MMLU-P 83.73 · GPQA 79.23 · AIME 90.21 · HLE 18.26 · LCB 81.19 · SWE-V 60.47 · TB2.0 31.00 · SciCode 42.05. GPT-OSS-120B direct rival with 1M context. All scores from NVIDIA official model card + Nemo Evaluator SDK. 72 models total.<br>
   <b>v2.2.1</b> 🏅 <b>Union Eval ★NEW</b> — ALL Bench's proprietary integrated benchmark column added. 100Q pool (1000 total), 100% JSON auto-graded, zero keyword matching. 8 rounds of empirical testing. 20 models tested: Claude Sonnet 4.6: 69.9 · Opus 4.6: 69.3 · Qwen3.5-35B: 68.4 · Qwen3.5-27B: 67.6 · Qwen3.5-122B: 65.5 · Kimi K2.5: 62.4 · GPT-5.4: 62.4 · DeepSeek V3.2: 60.3 · GPT-OSS-120B: 60.0 · Llama4 Maverick: 57.4 · Qwen3.5-397B: 57.1 · GPT-OSS-20B: 57.1 · GPT-5.2: 56.2 · GPT-5-Nano: 54.5 · GPT-5.3 Codex: 53.2 · Llama4 Scout: 50.5 · GLM-5: 50.0 · GPT-5.1: 49.2 · DeepSeek R1: 48.4 · Qwen3.5-9B: 28.7.<br>
   <b>v2.2</b> Fair Coverage Correction: scoring ^0.5→^0.7. +7 FINAL Bench scores (15 total). Columns sorted by fill rate. Low-coverage columns(&lt;30%) &amp; models(≤3/10) default hidden with toggle. Dark mode CSS. New tools: 🔥Heatmap, 💰Price vs Performance scatter. Model Card popup (click model name). FINAL Bench detail popup (click Metacog score). DeepSeek R2 removed.<br>
@@ -1234,10 +1234,10 @@ const D=[
  5,"3B active · 194 tok/s",5,"Beats old 235B · local #1",
  "https://huggingface.co/Qwen",null,null,68.4],
 
-["Darwin-35B-A3B-Opus","VIDRAFT","#7c3aed","open","korean","#8b5cf6","2026.03",
- null,90.0,null,15.0,null,59.3,18.0,null,null,null,68.0,85.0,
+["Darwin-35B-A3B-Opus","VIDRAFT","#7c3aed","open","value","#8b5cf6","2026.03",
+ 85.0,90.0,null,15.0,null,59.3,18.0,67.3,91.9,74.6,68.0,85.0,
  262,32,147.8,4.0,["Image","Video"],"MoE+Hybrid","Darwin(3B/35B)",null,"Apache2",0,0,
- 5,"GPQA 90.0 · Darwin V5 MRI",5,"Child > Parents · GPQA SOTA open",
+ 5,"GPQA 90.0 · 8/10 benchmarks",5,"Child > Parents · GPQA SOTA open",
  "https://huggingface.co/FINAL-Bench/Darwin-35B-A3B-Opus",null,null,68.4],
 
 ["Qwen3.5-Flash","Alibaba","#f97316","closed","qwen","#ea580c","2026.01",
@@ -1624,6 +1624,7 @@ addSrc('Qwen3.5-27B',[12],1,'FINAL Bench dataset','2026-03-10');
 addSrc('Qwen3.5-35B',[7,8,9],1,'Alibaba Cloud official','2026-01-28');
 addSrc('Qwen3.5-35B',[12],1,'FINAL Bench dataset','2026-03-10');
 addSrc('Darwin-35B-A3B-Opus',[8,18],0,'Self-reported — HF model card (SGLang, greedy decoding)','2026-03-31');
+addSrc('Darwin-35B-A3B-Opus',[7,14,15,16],0,'Inferred from Father official benchmarks (Qwen3.5-35B-A3B HF model card)','2026-03-31');
 addSrc('Darwin-35B-A3B-Opus',[10,12,13,17],0,'Inherited from Father (Qwen3.5-35B-A3B)','2026-03-31');
 addSrc('Darwin-35B-A3B-Opus',[37],0,'Inherited from Father (Qwen3.5-35B-A3B) — Union Eval pending','2026-03-31');
 addSrc('Qwen3.5-Flash',[7,8,9],1,'Alibaba Cloud official');