Update index.html

index.html

@@ -269,7 +269,7 @@ td.meta-col{background:rgba(99,102,241,.02);}
   <div class="li"><div class="ld" style="background:#e11d48"></div>C&lt;60%</div>
   <span style="color:#db2777;font-size:9.5px;margin-left:8px">★ = 신규지표</span>
   <span style="color:#16a34a;font-size:9.5px;margin-left:6px">💚 초록배경=오픈소스 가성비픽</span>
-  <span style="font-family:var(--font-mono);font-size:8.5px;color:#6366f1;margin-left:6px">🧬 Metacog = FINAL-Bench/Metacognitive (HF, 100문항)</span>
+  <span style="font-family:var(--font-mono);font-size:8.5px;color:#6366f1;margin-left:6px">🧬 Metacog = FINAL-Bench Baseline FINAL Score | 실측 9모델만 표기, 미평가 —</span>
 </div>
 </div>
 
@@ -319,7 +319,7 @@ td.meta-col{background:rgba(99,102,241,.02);}
 <!-- ========== TAB: INFO ========== -->
 <div id="info" class="tpane">
 <div class="info-grid" style="margin-top:14px">
-  <div class="fni"><b>🧬 Metacognitive ★v5.1 — FINAL-Bench</b><p>HuggingFace FINAL-Bench/Metacognitive 데이터셋. 100문항. 8가지 TICOS 유형: 자기수정(E)·함정탈출(A)·모순해결(B)·다중제약최적화(D)·점진적발견(C)·전문가패널(F)·피벗감지(G)·불확실성결정(H). A·B 등급 혼합. Claude 계열이 자기수정에 강점.</p></div>
+  <div class="fni"><b>🧬 Metacognitive ★v5.1 — FINAL-Bench (실측값)</b><p>FINAL-Bench/Metacognitive (HF 공식 리더보드). 9개 SOTA 모델 대상. Baseline FINAL Score(100점 만점) 기준. 🥇Kimi K2.5 <b>68.71</b> · 🥈GPT-5.2 <b>62.76</b> · 🥉GLM-4.7(GLM-5) <b>62.50</b> · Gemini 3.1 Pro <b>59.5</b> · Claude Opus 4.6 <b>56.04</b>(9위 최하). 나머지 모델은 미평가(—). 5축 평가: PQ·MA·ER·ID·FC. ER(자기교정)이 성능 향상의 94.8%를 담당.</p></div>
   <div class="fni"><b>🏆 종합점수 ★v5.1</b><p>MMLU-Pro·GPQA·AIME·HLE·Metacog·SWE-Pro·BFCL·IFEval·SWE-V 가용항목 단순평균. 미발표 제외. 전반적 AI 능력 종합지표.</p></div>
   <div class="fni"><b>🏭 공급사 ★v5.1</b><p>모델 개발사. OpenAI·Anthropic·Google·xAI·Alibaba·DeepSeek·Moonshot·Zhipu AI·Meta·Mistral. 공급사 철학과 강점이 모델 성격에 반영.</p></div>
   <div class="fni"><b>📅 출시일 ★v5.1</b><p>년도.월 기준. 2025.01(DeepSeek R1)~2026.02(GPT-5.2). 최신일수록 최신 학습데이터 및 아키텍처 개선 반영.</p></div>
@@ -348,57 +348,57 @@ td.meta-col{background:rgba(99,102,241,.02);}
 //  valScore(30), valLabel(31), csScore(32), csLabel(33)]
 const D=[
 ["GPT-5.2","OpenAI","#10a37f","closed","flagship","#10a37f","2026.02",
- 80.0,93.2,100,35.4,62,29.9,null,90.5,null,80.0,82,
+ 80.0,93.2,100,35.4,62.76,29.9,null,90.5,null,80.0,82,
  400,128,null,0.6,["이미지"],"Dense","Reasoning",1510,"Prop",1.75,14.0,
  3,"성능 최강급",4,"만능 선두·비용부담"],
 
 ["Claude Opus 4.6","Anthropic","#d97706","closed","flagship","#d97706","2025.10",
- 78.5,91.3,100,36.7,73,45.0,null,93.1,null,80.8,80,
+ 78.5,91.3,100,36.7,56.04,45.0,null,93.1,null,80.8,80,
  200,32,null,3.5,["이미지","영상"],"Dense","Reasoning(Adaptive)",1498,"Prop",5.0,25.0,
  3,"코딩·에이전트1위",5,"에이전트 최강·커뮤1위"],
 
 ["Gemini 3.1 Pro","Google","#4285f4","closed","flagship","#4285f4","2026.01",
- 83.0,94.3,97,44.7,61,43.3,null,91.0,null,74.2,88,
+ 83.0,94.3,97,44.7,59.5,43.3,null,91.0,null,74.2,88,
  2000,64,null,4.2,["이미지","영상","음성"],"Dense","Reasoning(DeepThink)",1501,"Prop",2.0,12.0,
  4,"GPQA·HLE·영상1위",4,"멀티모달 최강·HLE1위"],
 
 ["Grok 4 Heavy","xAI","#1d9bf0","closed","flagship","#1d9bf0","2025.11",
- 85.0,87.5,95,28.0,55,25.0,null,88.0,null,72.0,76,
+ 85.0,87.5,95,28.0,null,25.0,null,88.0,null,72.0,76,
  256,32,null,6.0,["이미지"],"Dense","Reasoning",1460,"Prop",3.0,15.0,
  3,"X·실시간검색",4,"수학·추론 강자"],
 
 ["Claude Sonnet 4.6","Anthropic","#d97706","closed","flagship","#f59e0b","2025.10",
- 76.0,84.5,93,22.0,68,43.6,null,92.0,null,72.7,79,
+ 76.0,84.5,93,22.0,null,43.6,null,92.0,null,72.7,79,
  200,64,null,2.0,["이미지","영상"],"Dense","Reasoning",1482,"Prop",3.0,15.0,
  4,"중급 최적선택",4,"실무 최다사용·밸런스"],
 
 ["GPT-OSS-120B","OpenAI","#059669","open","gptoss","#059669","2025.12",
- 90.0,80.9,97.9,12.0,52,16.2,null,null,null,null,72,
+ 90.0,80.9,97.9,12.0,null,16.2,null,null,null,null,72,
  128,32,null,null,["텍스트"],"MoE","Reasoning(5.1B/116.8B)",1380,"Apache2",0,0,
  5,"80GB 1장 로컬",4,"MMLU 오픈1위·o4-mini급"],
 
 ["GPT-OSS-20B","OpenAI","#059669","open","gptoss","#34d399","2025.12",
- 85.3,71.5,98.7,7.0,44,null,null,null,null,null,68,
+ 85.3,71.5,98.7,7.0,null,null,null,null,null,null,68,
  128,32,null,null,["텍스트"],"MoE","Reasoning(3.6B/20.9B)",1340,"Apache2",0,0,
  5,"16GB엣지·AIME1위",4,"소형 AIME 98.7% 최강"],
 
 ["Qwen3.5-397B","Alibaba","#f97316","open","qwen","#f97316","2026.01",
- 87.8,88.4,91.3,32.0,56,38.0,null,92.6,null,76.4,85,
+ 87.8,88.4,91.3,32.0,null,38.0,null,92.6,null,76.4,85,
  262,32,45,5.0,["이미지","영상"],"MoE+Hybrid","Reasoning(17B/397B)",1445,"Apache2",null,null,
  4,"오픈소스 플래그십",5,"IFBench 세계1위·201언어"],
 
 ["Qwen3.5-122B","Alibaba","#f97316","open","qwen","#fb923c","2026.01",
- null,87.6,85.0,24.0,50,28.0,72.2,93.4,null,72.2,82,
+ null,87.6,85.0,24.0,null,28.0,72.2,93.4,null,72.2,82,
  262,32,null,6.0,["이미지","영상"],"MoE+Hybrid","Reasoning(10B/122B)",1420,"Apache2",0.40,1.20,
  5,"BFCL 세계1위",5,"GPT-5 mini 대비 BFCL+30%"],
 
 ["Qwen3.5-27B","Alibaba","#f97316","open","qwen","#fdba74","2026.01",
- 86.1,85.5,null,18.0,47,20.0,null,null,null,72.4,80,
+ 86.1,85.5,null,18.0,null,20.0,null,null,null,72.4,80,
  262,32,null,5.5,["이미지","영상"],"Dense","Reasoning(27B)",1395,"Apache2",null,null,
  5,"Dense·코딩 특화",4,"SWE 72.4%·GPT-5 mini동급"],
 
 ["Qwen3.5-35B","Alibaba","#f97316","open","qwen","#fed7aa","2026.01",
- null,83.0,null,15.0,43,18.0,null,null,null,68.0,78,
+ null,83.0,null,15.0,null,18.0,null,null,null,68.0,78,
  262,32,194,4.0,["이미지","영상"],"MoE+Hybrid","Reasoning(3B/35B)",1380,"Apache2",0.10,0.40,
  5,"3B활성·194tok/s",5,"이전 235B 능가·로컬1위"],
 
@@ -408,7 +408,7 @@ const D=[
  5,"초저가·1M ctx",4,"DeepSeek보다 저렴+멀티모달"],
 
 ["Qwen3.5-9B","Alibaba","#f97316","open","qwen","#c2410c","2026.01",
- null,81.7,null,10.0,40,null,null,null,null,null,73,
+ null,81.7,null,10.0,null,null,null,null,null,null,73,
  262,32,null,3.5,["이미지","영상"],"Dense","Reasoning(9B)",1300,"Apache2",null,null,
  5,"9B로 120B 격파",5,"GPQA 81.7>80.9·RTX3060"],
 
@@ -418,32 +418,32 @@ const D=[
  5,"최초 4B 멀티모달",4,"영상처리·최소 에이전트"],
 
 ["DeepSeek V3.2","DeepSeek","#6366f1","open","value","#6366f1","2025.12",
- null,79.9,89.3,22.0,48,15.6,null,87.5,null,66.0,78,
+ null,79.9,89.3,22.0,null,15.6,null,87.5,null,66.0,78,
  128,8,null,7.0,["텍스트"],"MoE","Non-Reasoning(37B/671B)",1430,"MIT",0.14,0.28,
  5,"MIT·가성비 레전드",5,"레딧 최애·가성비 압도적"],
 
 ["DeepSeek R1","DeepSeek","#6366f1","open","value","#818cf8","2025.01",
- 85.0,82.0,87.5,14.0,53,18.0,null,83.3,null,49.2,74,
+ 85.0,82.0,87.5,14.0,null,18.0,null,83.3,null,49.2,74,
  128,8,null,8.0,["텍스트"],"MoE","Reasoning(37B/671B)",1440,"MIT",0.55,2.19,
  5,"추론 특화",5,"수학·추론 전설·레딧 최애"],
 
 ["Kimi K2.5","Moonshot","#8b5cf6","open","flagship","#8b5cf6","2025.11",
- 92.0,87.6,96.1,44.9,59,27.7,null,94.0,null,76.8,81,
+ 92.0,87.6,96.1,44.9,68.71,27.7,null,94.0,null,76.8,81,
  200,32,null,5.0,["이미지","영상"],"MoE","Reasoning(1T)",1447,"MIT",0.55,2.50,
  5,"HLE·MMLU 최강급",4,"HLE 44.9% 1위급·에이전트"],
 
 ["GLM-4.7","Zhipu AI","#14b8a6","open","value","#14b8a6","2025.12",
- 84.3,85.7,95.7,21.0,51,null,null,88.0,89.0,77.8,74,
+ 84.3,85.7,95.7,21.0,62.50,null,null,88.0,89.0,77.8,74,
  128,16,null,3.0,["이미지"],"Dense","Reasoning",1445,"MIT",0.35,0.39,
  5,"MIT·초저가·ELO1위",4,"Arena ELO 1445·SWE 최상위"],
 
 ["Llama 4 Scout","Meta","#0081fb","open","flagship","#0081fb","2025.04",
- null,73.0,85.0,12.0,39,5.2,null,85.0,null,55.0,70,
+ null,73.0,85.0,12.0,null,5.2,null,85.0,null,55.0,70,
  10000,16,null,2.0,["이미지","영상"],"MoE","Non-Reasoning(17B/400B)",1340,"Meta",0.11,0.34,
  4,"10M ctx 혁명",3,"로컬 선호·10M 컨텍스트"],
 
 ["Mistral Large 3","Mistral","#ff7043","open","flagship","#ff7043","2025.11",
- null,78.0,82.0,11.0,42,12.0,null,86.0,null,60.0,72,
+ null,78.0,82.0,11.0,null,12.0,null,86.0,null,60.0,72,
  256,16,null,0.3,["이미지"],"MoE","Non-Reasoning(675B)",1320,"Apache2",2.0,6.0,
  3,"TTFT 0.3s 최속",3,"GDPR준수·유럽 선호"]
 ];