<!DOCTYPE html>
<html lang="zh-CN">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>DeepSeek V4 发布 — 迈向百万 Token 高效智能</title>
<link rel="preconnect" href="https://fonts.googleapis.com">
<link href="https://fonts.googleapis.com/css2?family=Space+Grotesk:wght@300;400;500;600;700&family=Noto+Sans+SC:wght@300;400;500;700;900&family=JetBrains+Mono:wght@400;500;600&display=swap" rel="stylesheet">
<style>
:root {
  --bg-primary: #0a0a0f;
  --bg-secondary: #111118;
  --bg-card: #16161f;
  --bg-card-hover: #1c1c28;
  --text-primary: #e8e8f0;
  --text-secondary: #9898b0;
  --text-muted: #6868880;
  --accent-blue: #4f8cff;
  --accent-cyan: #00d4ff;
  --accent-purple: #8b5cf6;
  --accent-green: #10b981;
  --accent-orange: #f59e0b;
  --accent-pink: #ec4899;
  --accent-red: #ef4444;
  --gradient-hero: linear-gradient(135deg, #4f8cff 0%, #8b5cf6 50%, #ec4899 100%);
  --gradient-card: linear-gradient(145deg, rgba(79,140,255,0.08), rgba(139,92,246,0.08));
  --border-subtle: rgba(255,255,255,0.06);
  --border-accent: rgba(79,140,255,0.3);
  --radius-sm: 8px;
  --radius-md: 12px;
  --radius-lg: 20px;
  --radius-xl: 28px;
  --font-heading: 'Space Grotesk', 'Noto Sans SC', sans-serif;
  --font-body: 'Noto Sans SC', 'Space Grotesk', sans-serif;
  --font-mono: 'JetBrains Mono', monospace;
}

* { margin: 0; padding: 0; box-sizing: border-box; }

html {
  scroll-behavior: smooth;
  font-size: 16px;
}

body {
  font-family: var(--font-body);
  background: var(--bg-primary);
  color: var(--text-primary);
  line-height: 1.7;
  overflow-x: hidden;
}

/* Noise texture overlay */
body::before {
  content: '';
  position: fixed;
  inset: 0;
  background-image: url("data:image/svg+xml,%3Csvg viewBox='0 0 256 256' xmlns='http://www.w3.org/2000/svg'%3E%3Cfilter id='n'%3E%3CfeTurbulence type='fractalNoise' baseFrequency='0.9' numOctaves='4' /%3E%3C/filter%3E%3Crect width='100%25' height='100%25' filter='url(%23n)' opacity='0.03'/%3E%3C/svg%3E");
  pointer-events: none;
  z-index: 0;
}

.container {
  max-width: 1100px;
  margin: 0 auto;
  padding: 0 24px;
  position: relative;
  z-index: 1;
}

/* ====== HERO ====== */
.hero {
  position: relative;
  min-height: 100vh;
  display: flex;
  flex-direction: column;
  justify-content: center;
  align-items: center;
  text-align: center;
  padding: 80px 24px;
  overflow: hidden;
}

.hero-bg {
  position: absolute;
  inset: 0;
  background: radial-gradient(ellipse 80% 60% at 50% 40%, rgba(79,140,255,0.12) 0%, transparent 60%),
              radial-gradient(ellipse 60% 50% at 30% 70%, rgba(139,92,246,0.08) 0%, transparent 50%),
              radial-gradient(ellipse 50% 40% at 80% 20%, rgba(236,72,153,0.06) 0%, transparent 50%);
  z-index: -1;
}

.hero-grid {
  position: absolute;
  inset: 0;
  background: linear-gradient(rgba(255,255,255,0.02) 1px, transparent 1px),
              linear-gradient(90deg, rgba(255,255,255,0.02) 1px, transparent 1px);
  background-size: 60px 60px;
  mask-image: radial-gradient(ellipse 70% 60% at 50% 50%, black 20%, transparent 70%);
  z-index: -1;
}

.hero-badge {
  display: inline-flex;
  align-items: center;
  gap: 8px;
  padding: 6px 18px;
  border: 1px solid var(--border-accent);
  border-radius: 100px;
  background: rgba(79,140,255,0.08);
  font-family: var(--font-mono);
  font-size: 0.8rem;
  color: var(--accent-blue);
  letter-spacing: 0.05em;
  margin-bottom: 32px;
  animation: fadeInDown 0.8s ease-out;
}

.hero-badge .dot {
  width: 6px; height: 6px;
  background: var(--accent-green);
  border-radius: 50%;
  animation: pulse 2s ease-in-out infinite;
}

@keyframes pulse {
  0%, 100% { opacity: 1; }
  50% { opacity: 0.4; }
}

.hero h1 {
  font-family: var(--font-heading);
  font-size: clamp(2.8rem, 6vw, 5.5rem);
  font-weight: 700;
  line-height: 1.1;
  letter-spacing: -0.03em;
  margin-bottom: 24px;
  animation: fadeInUp 0.8s ease-out 0.1s both;
}

.hero h1 .gradient-text {
  background: var(--gradient-hero);
  -webkit-background-clip: text;
  -webkit-text-fill-color: transparent;
  background-clip: text;
}

.hero .subtitle {
  font-size: clamp(1.1rem, 2vw, 1.35rem);
  color: var(--text-secondary);
  max-width: 680px;
  margin: 0 auto 48px;
  font-weight: 300;
  animation: fadeInUp 0.8s ease-out 0.25s both;
}

.hero-stats {
  display: grid;
  grid-template-columns: repeat(4, 1fr);
  gap: 24px;
  max-width: 800px;
  width: 100%;
  animation: fadeInUp 0.8s ease-out 0.4s both;
}

.hero-stat {
  padding: 24px 16px;
  border-radius: var(--radius-lg);
  background: rgba(255,255,255,0.03);
  border: 1px solid var(--border-subtle);
  backdrop-filter: blur(8px);
  transition: all 0.3s ease;
}

.hero-stat:hover {
  background: rgba(255,255,255,0.06);
  border-color: var(--border-accent);
  transform: translateY(-4px);
}

.hero-stat .value {
  font-family: var(--font-heading);
  font-size: 2rem;
  font-weight: 700;
  color: var(--accent-blue);
  margin-bottom: 4px;
}

.hero-stat .label {
  font-size: 0.82rem;
  color: var(--text-secondary);
  font-weight: 400;
}

@keyframes fadeInDown {
  from { opacity: 0; transform: translateY(-20px); }
  to { opacity: 1; transform: translateY(0); }
}
@keyframes fadeInUp {
  from { opacity: 0; transform: translateY(30px); }
  to { opacity: 1; transform: translateY(0); }
}

/* ====== SECTIONS ====== */
section {
  padding: 100px 0;
}

.section-label {
  display: inline-flex;
  align-items: center;
  gap: 8px;
  font-family: var(--font-mono);
  font-size: 0.75rem;
  color: var(--accent-blue);
  text-transform: uppercase;
  letter-spacing: 0.15em;
  margin-bottom: 16px;
}

.section-label::before {
  content: '';
  width: 24px;
  height: 1px;
  background: var(--accent-blue);
}

.section-title {
  font-family: var(--font-heading);
  font-size: clamp(2rem, 4vw, 3rem);
  font-weight: 700;
  letter-spacing: -0.02em;
  line-height: 1.2;
  margin-bottom: 20px;
}

.section-desc {
  color: var(--text-secondary);
  font-size: 1.05rem;
  max-width: 640px;
  margin-bottom: 48px;
  font-weight: 300;
}

/* ====== MODEL CARDS ====== */
.model-cards {
  display: grid;
  grid-template-columns: 1fr 1fr;
  gap: 24px;
  margin-bottom: 48px;
}

.model-card {
  border-radius: var(--radius-lg);
  padding: 36px;
  border: 1px solid var(--border-subtle);
  position: relative;
  overflow: hidden;
  transition: all 0.4s ease;
}

.model-card::before {
  content: '';
  position: absolute;
  inset: 0;
  opacity: 0;
  transition: opacity 0.4s ease;
}

.model-card:hover {
  border-color: var(--border-accent);
  transform: translateY(-4px);
}

.model-card:hover::before {
  opacity: 1;
}

.model-card.pro {
  background: linear-gradient(145deg, rgba(79,140,255,0.06), rgba(139,92,246,0.04));
}
.model-card.pro::before {
  background: linear-gradient(145deg, rgba(79,140,255,0.1), rgba(139,92,246,0.06));
}

.model-card.flash {
  background: linear-gradient(145deg, rgba(0,212,255,0.06), rgba(16,185,129,0.04));
}
.model-card.flash::before {
  background: linear-gradient(145deg, rgba(0,212,255,0.1), rgba(16,185,129,0.06));
}

.model-card .badge {
  display: inline-block;
  padding: 4px 12px;
  border-radius: 100px;
  font-family: var(--font-mono);
  font-size: 0.7rem;
  font-weight: 600;
  letter-spacing: 0.08em;
  text-transform: uppercase;
  margin-bottom: 16px;
}

.model-card.pro .badge {
  background: rgba(79,140,255,0.15);
  color: var(--accent-blue);
}

.model-card.flash .badge {
  background: rgba(0,212,255,0.15);
  color: var(--accent-cyan);
}

.model-card h3 {
  font-family: var(--font-heading);
  font-size: 1.6rem;
  font-weight: 600;
  margin-bottom: 12px;
}

.model-card p {
  color: var(--text-secondary);
  font-size: 0.92rem;
  margin-bottom: 24px;
  line-height: 1.7;
}

.model-specs {
  display: grid;
  grid-template-columns: 1fr 1fr;
  gap: 12px;
}

.spec-item {
  padding: 12px 16px;
  border-radius: var(--radius-sm);
  background: rgba(0,0,0,0.3);
}

.spec-item .spec-label {
  font-size: 0.72rem;
  color: var(--text-secondary);
  text-transform: uppercase;
  letter-spacing: 0.08em;
  font-family: var(--font-mono);
  margin-bottom: 4px;
}

.spec-item .spec-value {
  font-family: var(--font-heading);
  font-size: 1.1rem;
  font-weight: 600;
}

/* ====== ARCHITECTURE ====== */
.arch-features {
  display: grid;
  grid-template-columns: repeat(3, 1fr);
  gap: 20px;
  margin-top: 48px;
}

.arch-card {
  padding: 32px 28px;
  border-radius: var(--radius-lg);
  background: var(--bg-card);
  border: 1px solid var(--border-subtle);
  transition: all 0.3s ease;
  position: relative;
}

.arch-card:hover {
  background: var(--bg-card-hover);
  border-color: var(--border-accent);
  transform: translateY(-4px);
}

.arch-icon {
  width: 48px; height: 48px;
  border-radius: var(--radius-md);
  display: flex;
  align-items: center;
  justify-content: center;
  font-size: 1.4rem;
  margin-bottom: 20px;
}

.arch-card:nth-child(1) .arch-icon { background: rgba(79,140,255,0.12); }
.arch-card:nth-child(2) .arch-icon { background: rgba(139,92,246,0.12); }
.arch-card:nth-child(3) .arch-icon { background: rgba(16,185,129,0.12); }

.arch-card h4 {
  font-family: var(--font-heading);
  font-size: 1.1rem;
  font-weight: 600;
  margin-bottom: 10px;
}

.arch-card p {
  color: var(--text-secondary);
  font-size: 0.88rem;
  line-height: 1.7;
}

/* ====== BENCHMARK CHART ====== */
.chart-section {
  background: var(--bg-secondary);
  border-radius: var(--radius-xl);
  border: 1px solid var(--border-subtle);
  padding: 48px;
  margin: 48px 0;
}

.chart-header {
  display: flex;
  justify-content: space-between;
  align-items: flex-end;
  margin-bottom: 40px;
  flex-wrap: wrap;
  gap: 16px;
}

.chart-legend {
  display: flex;
  gap: 20px;
  flex-wrap: wrap;
}

.legend-item {
  display: flex;
  align-items: center;
  gap: 8px;
  font-size: 0.82rem;
  color: var(--text-secondary);
}

.legend-dot {
  width: 10px; height: 10px;
  border-radius: 3px;
}

/* Bar chart */
.bar-chart {
  width: 100%;
}

.bar-group {
  display: grid;
  grid-template-columns: 160px 1fr;
  gap: 16px;
  align-items: center;
  margin-bottom: 14px;
}

.bar-label {
  font-size: 0.82rem;
  color: var(--text-secondary);
  text-align: right;
  font-family: var(--font-mono);
  white-space: nowrap;
}

.bar-tracks {
  display: flex;
  flex-direction: column;
  gap: 4px;
}

.bar-track {
  height: 22px;
  border-radius: 4px;
  background: rgba(255,255,255,0.03);
  position: relative;
  overflow: hidden;
}

.bar-fill {
  height: 100%;
  border-radius: 4px;
  position: relative;
  transition: width 1.5s cubic-bezier(0.22, 0.61, 0.36, 1);
  display: flex;
  align-items: center;
  justify-content: flex-end;
  padding-right: 8px;
}

.bar-fill span {
  font-size: 0.68rem;
  font-family: var(--font-mono);
  font-weight: 600;
  color: rgba(255,255,255,0.9);
  white-space: nowrap;
}

.bar-fill.v4pro { background: linear-gradient(90deg, rgba(79,140,255,0.7), var(--accent-blue)); }
.bar-fill.v4flash { background: linear-gradient(90deg, rgba(0,212,255,0.6), var(--accent-cyan)); }
.bar-fill.v3 { background: linear-gradient(90deg, rgba(139,92,246,0.5), var(--accent-purple)); }

/* ====== RADAR / Comparison table ====== */
.comparison-table {
  width: 100%;
  border-collapse: separate;
  border-spacing: 0;
  font-size: 0.85rem;
}

.comparison-table thead th {
  padding: 14px 16px;
  text-align: center;
  font-family: var(--font-mono);
  font-size: 0.72rem;
  font-weight: 600;
  text-transform: uppercase;
  letter-spacing: 0.08em;
  color: var(--text-secondary);
  border-bottom: 1px solid var(--border-subtle);
}

.comparison-table thead th:first-child {
  text-align: left;
}

.comparison-table tbody td {
  padding: 12px 16px;
  text-align: center;
  border-bottom: 1px solid rgba(255,255,255,0.03);
  font-family: var(--font-mono);
  font-size: 0.82rem;
}

.comparison-table tbody td:first-child {
  text-align: left;
  font-family: var(--font-body);
  color: var(--text-secondary);
}

.comparison-table tbody tr:hover {
  background: rgba(255,255,255,0.02);
}

.highlight-cell {
  color: var(--accent-blue) !important;
  font-weight: 600;
}

.best-cell {
  background: rgba(79,140,255,0.08);
  color: var(--accent-blue) !important;
  font-weight: 700;
  border-radius: 4px;
}

/* ====== EFFICIENCY SECTION ====== */
.efficiency-grid {
  display: grid;
  grid-template-columns: repeat(3, 1fr);
  gap: 24px;
  margin-top: 48px;
}

.eff-card {
  padding: 32px;
  border-radius: var(--radius-lg);
  background: var(--bg-card);
  border: 1px solid var(--border-subtle);
  text-align: center;
  transition: all 0.3s ease;
}

.eff-card:hover {
  transform: translateY(-4px);
  border-color: var(--border-accent);
}

.eff-ring {
  width: 120px; height: 120px;
  margin: 0 auto 20px;
  position: relative;
}

.eff-ring svg {
  width: 100%; height: 100%;
  transform: rotate(-90deg);
}

.eff-ring circle {
  fill: none;
  stroke-width: 6;
}

.eff-ring .bg { stroke: rgba(255,255,255,0.05); }
.eff-ring .progress { stroke-linecap: round; transition: stroke-dashoffset 2s ease; }

.eff-ring .center-text {
  position: absolute;
  inset: 0;
  display: flex;
  flex-direction: column;
  align-items: center;
  justify-content: center;
}

.eff-ring .center-text .num {
  font-family: var(--font-heading);
  font-size: 1.5rem;
  font-weight: 700;
}

.eff-ring .center-text .unit {
  font-size: 0.7rem;
  color: var(--text-secondary);
}

.eff-card h4 {
  font-family: var(--font-heading);
  font-size: 1rem;
  margin-bottom: 8px;
}

.eff-card p {
  font-size: 0.82rem;
  color: var(--text-secondary);
  line-height: 1.6;
}

/* ====== REASONING MODES ====== */
.modes-grid {
  display: grid;
  grid-template-columns: repeat(3, 1fr);
  gap: 20px;
  margin-top: 48px;
}

.mode-card {
  padding: 32px 28px;
  border-radius: var(--radius-lg);
  border: 1px solid var(--border-subtle);
  background: var(--bg-card);
  transition: all 0.3s ease;
}

.mode-card:hover {
  transform: translateY(-4px);
  border-color: var(--border-accent);
}

.mode-icon {
  font-size: 2rem;
  margin-bottom: 16px;
}

.mode-card h4 {
  font-family: var(--font-heading);
  font-size: 1.1rem;
  margin-bottom: 8px;
}

.mode-card .mode-tag {
  display: inline-block;
  font-family: var(--font-mono);
  font-size: 0.7rem;
  padding: 3px 10px;
  border-radius: 100px;
  margin-bottom: 12px;
}

.mode-card:nth-child(1) .mode-tag { background: rgba(16,185,129,0.12); color: var(--accent-green); }
.mode-card:nth-child(2) .mode-tag { background: rgba(245,158,11,0.12); color: var(--accent-orange); }
.mode-card:nth-child(3) .mode-tag { background: rgba(236,72,153,0.12); color: var(--accent-pink); }

.mode-card p {
  font-size: 0.88rem;
  color: var(--text-secondary);
  line-height: 1.7;
}

/* ====== TRAINING ====== */
.training-stats {
  display: grid;
  grid-template-columns: repeat(4, 1fr);
  gap: 20px;
  margin-top: 48px;
}

.train-stat {
  padding: 28px 24px;
  border-radius: var(--radius-lg);
  background: var(--bg-card);
  border: 1px solid var(--border-subtle);
  text-align: center;
  transition: all 0.3s ease;
}

.train-stat:hover {
  transform: translateY(-4px);
  border-color: var(--border-accent);
}

.train-stat .num {
  font-family: var(--font-heading);
  font-size: 2.2rem;
  font-weight: 700;
  margin-bottom: 4px;
}

.train-stat:nth-child(1) .num { color: var(--accent-blue); }
.train-stat:nth-child(2) .num { color: var(--accent-purple); }
.train-stat:nth-child(3) .num { color: var(--accent-cyan); }
.train-stat:nth-child(4) .num { color: var(--accent-green); }

.train-stat .desc {
  font-size: 0.82rem;
  color: var(--text-secondary);
}

/* ====== TIMELINE ====== */
.timeline {
  position: relative;
  margin: 48px 0;
  padding-left: 48px;
}

.timeline::before {
  content: '';
  position: absolute;
  left: 16px;
  top: 0;
  bottom: 0;
  width: 2px;
  background: linear-gradient(to bottom, var(--accent-blue), var(--accent-purple), var(--accent-pink));
  border-radius: 2px;
}

.timeline-item {
  position: relative;
  margin-bottom: 36px;
  padding: 24px 28px;
  background: var(--bg-card);
  border: 1px solid var(--border-subtle);
  border-radius: var(--radius-md);
  transition: all 0.3s ease;
}

.timeline-item:hover {
  border-color: var(--border-accent);
  transform: translateX(4px);
}

.timeline-item::before {
  content: '';
  position: absolute;
  left: -40px;
  top: 30px;
  width: 12px; height: 12px;
  background: var(--accent-blue);
  border-radius: 50%;
  border: 3px solid var(--bg-primary);
}

.timeline-item .year {
  font-family: var(--font-mono);
  font-size: 0.75rem;
  color: var(--accent-blue);
  margin-bottom: 8px;
  letter-spacing: 0.05em;
}

.timeline-item h4 {
  font-family: var(--font-heading);
  font-size: 1.05rem;
  margin-bottom: 8px;
}

.timeline-item p {
  font-size: 0.85rem;
  color: var(--text-secondary);
  line-height: 1.7;
}

/* ====== FOOTER ====== */
.footer {
  padding: 60px 0;
  border-top: 1px solid var(--border-subtle);
  text-align: center;
}

.footer p {
  font-size: 0.85rem;
  color: var(--text-secondary);
}

.footer .links {
  display: flex;
  justify-content: center;
  gap: 24px;
  margin-top: 20px;
}

.footer .links a {
  color: var(--accent-blue);
  text-decoration: none;
  font-size: 0.85rem;
  font-family: var(--font-mono);
  transition: color 0.2s;
}

.footer .links a:hover {
  color: var(--accent-cyan);
}

/* ====== SCROLL ANIMATION ====== */
.reveal {
  opacity: 0;
  transform: translateY(40px);
  transition: all 0.8s cubic-bezier(0.22, 0.61, 0.36, 1);
}

.reveal.visible {
  opacity: 1;
  transform: translateY(0);
}

/* ====== RESPONSIVE ====== */
@media (max-width: 900px) {
  .hero-stats { grid-template-columns: repeat(2, 1fr); }
  .model-cards { grid-template-columns: 1fr; }
  .arch-features, .efficiency-grid, .modes-grid { grid-template-columns: 1fr; }
  .training-stats { grid-template-columns: repeat(2, 1fr); }
  .bar-group { grid-template-columns: 100px 1fr; }
  .chart-section { padding: 28px 20px; }
  .comparison-table { font-size: 0.75rem; }
}

@media (max-width: 600px) {
  .hero-stats { grid-template-columns: 1fr 1fr; gap: 12px; }
  .hero-stat .value { font-size: 1.4rem; }
  .training-stats { grid-template-columns: 1fr 1fr; }
  section { padding: 60px 0; }
}

/* ====== DIVIDER ====== */
.section-divider {
  height: 1px;
  background: linear-gradient(90deg, transparent, var(--border-accent), transparent);
  margin: 0;
}

/* Scrollbar */
::-webkit-scrollbar { width: 6px; }
::-webkit-scrollbar-track { background: var(--bg-primary); }
::-webkit-scrollbar-thumb { background: rgba(255,255,255,0.1); border-radius: 3px; }
::-webkit-scrollbar-thumb:hover { background: rgba(255,255,255,0.2); }
</style>
</head>
<body>

<!-- ====== HERO ====== -->
<header class="hero">
  <div class="hero-bg"></div>
  <div class="hero-grid"></div>
  
  <div class="hero-badge">
    <span class="dot"></span>
    2026.04.24 — 全球发布
  </div>
  
  <h1>
    <span class="gradient-text">DeepSeek V4</span><br>
    百万 Token 高效智能新纪元
  </h1>
  
  <p class="subtitle">
    全新 MoE 架构，1.6T 参数旗舰 Pro 与轻量 Flash 双模型齐发。<br>
    原生百万 Token 上下文，仅需 V3 系列 27% 推理算力和 10% KV 缓存。
  </p>
  
  <div class="hero-stats">
    <div class="hero-stat">
      <div class="value">1.6T</div>
      <div class="label">Pro 总参数</div>
    </div>
    <div class="hero-stat">
      <div class="value">1M</div>
      <div class="label">上下文长度</div>
    </div>
    <div class="hero-stat">
      <div class="value">32T+</div>
      <div class="label">预训练 Tokens</div>
    </div>
    <div class="hero-stat">
      <div class="value">27%</div>
      <div class="label">推理 FLOPs (vs V3)</div>
    </div>
  </div>
</header>

<!-- ====== MODEL OVERVIEW ====== -->
<div class="section-divider"></div>
<section>
  <div class="container">
    <div class="reveal">
      <div class="section-label">Model Overview</div>
      <h2 class="section-title">双模型矩阵：Pro × Flash</h2>
      <p class="section-desc">
        DeepSeek-V4 同时发布 Pro 和 Flash 两个版本，覆盖从极致性能到高效部署的全场景需求。
      </p>
    </div>
    
    <div class="model-cards reveal">
      <div class="model-card pro">
        <span class="badge">旗舰级</span>
        <h3>DeepSeek-V4-Pro</h3>
        <p>1.6 万亿参数的 MoE 旗舰模型，每次推理激活 49B 参数。在知识、推理、编程、数学和智能体任务上全面对标最强闭源模型。</p>
        <div class="model-specs">
          <div class="spec-item">
            <div class="spec-label">总参数</div>
            <div class="spec-value">1.6T</div>
          </div>
          <div class="spec-item">
            <div class="spec-label">激活参数</div>
            <div class="spec-value">49B</div>
          </div>
          <div class="spec-item">
            <div class="spec-label">上下文</div>
            <div class="spec-value">1M tokens</div>
          </div>
          <div class="spec-item">
            <div class="spec-label">精度</div>
            <div class="spec-value">FP4+FP8</div>
          </div>
        </div>
      </div>
      
      <div class="model-card flash">
        <span class="badge">效率之王</span>
        <h3>DeepSeek-V4-Flash</h3>
        <p>284B 参数的轻量级 MoE 模型，每次推理仅激活 13B 参数。在 Max 推理模式下可逼近 Pro 版本的推理能力，极致性价比。</p>
        <div class="model-specs">
          <div class="spec-item">
            <div class="spec-label">总参数</div>
            <div class="spec-value">284B</div>
          </div>
          <div class="spec-item">
            <div class="spec-label">激活参数</div>
            <div class="spec-value">13B</div>
          </div>
          <div class="spec-item">
            <div class="spec-label">上下文</div>
            <div class="spec-value">1M tokens</div>
          </div>
          <div class="spec-item">
            <div class="spec-label">精度</div>
            <div class="spec-value">FP4+FP8</div>
          </div>
        </div>
      </div>
    </div>
  </div>
</section>

<!-- ====== ARCHITECTURE ====== -->
<div class="section-divider"></div>
<section>
  <div class="container">
    <div class="reveal">
      <div class="section-label">Architecture</div>
      <h2 class="section-title">三大核心架构创新</h2>
      <p class="section-desc">
        DeepSeek-V4 在注意力机制、信号传播和训练优化器三个方面实现了突破性创新，大幅提升长上下文处理效率。
      </p>
    </div>
    
    <div class="arch-features reveal">
      <div class="arch-card">
        <div class="arch-icon">🧩</div>
        <h4>混合注意力架构<br><small style="color:var(--text-secondary);font-weight:400;font-size:0.75rem">CSA + HCA</small></h4>
        <p>将压缩稀疏注意力（CSA）与重度压缩注意力（HCA）巧妙融合。在百万 Token 上下文设置下，单 Token 推理 FLOPs 仅为 V3 的 27%，KV 缓存减少至 10%。</p>
      </div>
      <div class="arch-card">
        <div class="arch-icon">🌐</div>
        <h4>流形约束超连接<br><small style="color:var(--text-secondary);font-weight:400;font-size:0.75rem">mHC</small></h4>
        <p>引入流形约束超连接（Manifold-Constrained Hyper-Connections），增强信号在深层网络中的传播稳定性，同时保持模型的表达能力。</p>
      </div>
      <div class="arch-card">
        <div class="arch-icon">⚡</div>
        <h4>Muon 优化器<br><small style="color:var(--text-secondary);font-weight:400;font-size:0.75rem">Faster Convergence</small></h4>
        <p>采用全新 Muon 优化器替代传统 Adam，实现更快的训练收敛速度和更出色的训练稳定性，显著降低训练成本。</p>
      </div>
    </div>
  </div>
</section>

<!-- ====== EFFICIENCY ====== -->
<div class="section-divider"></div>
<section>
  <div class="container">
    <div class="reveal">
      <div class="section-label">Efficiency</div>
      <h2 class="section-title">效率革命：相比 V3 的飞跃</h2>
      <p class="section-desc">
        在支持百万 Token 上下文的同时，DeepSeek-V4 将推理成本大幅降低，真正实现了"更强且更高效"。
      </p>
    </div>
    
    <div class="efficiency-grid reveal">
      <div class="eff-card">
        <div class="eff-ring">
          <svg viewBox="0 0 120 120">
            <circle class="bg" cx="60" cy="60" r="52" />
            <circle class="progress" cx="60" cy="60" r="52" stroke="var(--accent-blue)"
              stroke-dasharray="327" stroke-dashoffset="238" />
          </svg>
          <div class="center-text">
            <span class="num" style="color:var(--accent-blue)">27%</span>
            <span class="unit">of V3.2</span>
          </div>
        </div>
        <h4>推理 FLOPs</h4>
        <p>百万 Token 场景下，单 Token 推理计算量仅为 V3.2 的 27%</p>
      </div>
      
      <div class="eff-card">
        <div class="eff-ring">
          <svg viewBox="0 0 120 120">
            <circle class="bg" cx="60" cy="60" r="52" />
            <circle class="progress" cx="60" cy="60" r="52" stroke="var(--accent-purple)"
              stroke-dasharray="327" stroke-dashoffset="294" />
          </svg>
          <div class="center-text">
            <span class="num" style="color:var(--accent-purple)">10%</span>
            <span class="unit">of V3.2</span>
          </div>
        </div>
        <h4>KV 缓存</h4>
        <p>百万 Token 上下文中 KV 缓存用量仅为 V3.2 的 10%，显存需求大幅降低</p>
      </div>
      
      <div class="eff-card">
        <div class="eff-ring">
          <svg viewBox="0 0 120 120">
            <circle class="bg" cx="60" cy="60" r="52" />
            <circle class="progress" cx="60" cy="60" r="52" stroke="var(--accent-green)"
              stroke-dasharray="327" stroke-dashoffset="0" />
          </svg>
          <div class="center-text">
            <span class="num" style="color:var(--accent-green)">1M</span>
            <span class="unit">tokens</span>
          </div>
        </div>
        <h4>原生上下文</h4>
        <p>原生支持百万 Token 上下文窗口，无需外部扩展即可处理超长文档</p>
      </div>
    </div>
  </div>
</section>

<!-- ====== BENCHMARK ====== -->
<div class="section-divider"></div>
<section>
  <div class="container">
    <div class="reveal">
      <div class="section-label">Benchmarks</div>
      <h2 class="section-title">基准测试：Base 模型对比</h2>
      <p class="section-desc">
        V4-Pro-Base 在绝大多数评测中领先 V3.2-Base，展示了架构升级带来的全面提升。
      </p>
    </div>
    
    <div class="chart-section reveal">
      <div class="chart-header">
        <h3 style="font-family:var(--font-heading);font-size:1.2rem;">核心基准得分对比</h3>
        <div class="chart-legend">
          <div class="legend-item"><div class="legend-dot" style="background:var(--accent-blue)"></div>V4-Pro</div>
          <div class="legend-item"><div class="legend-dot" style="background:var(--accent-cyan)"></div>V4-Flash</div>
          <div class="legend-item"><div class="legend-dot" style="background:var(--accent-purple)"></div>V3.2</div>
        </div>
      </div>
      
      <div class="bar-chart">
        <div class="bar-group">
          <div class="bar-label">MMLU</div>
          <div class="bar-tracks">
            <div class="bar-track"><div class="bar-fill v4pro" style="width:90.1%"><span>90.1</span></div></div>
            <div class="bar-track"><div class="bar-fill v4flash" style="width:88.7%"><span>88.7</span></div></div>
            <div class="bar-track"><div class="bar-fill v3" style="width:87.8%"><span>87.8</span></div></div>
          </div>
        </div>
        
        <div class="bar-group">
          <div class="bar-label">MMLU-Pro</div>
          <div class="bar-tracks">
            <div class="bar-track"><div class="bar-fill v4pro" style="width:73.5%"><span>73.5</span></div></div>
            <div class="bar-track"><div class="bar-fill v4flash" style="width:68.3%"><span>68.3</span></div></div>
            <div class="bar-track"><div class="bar-fill v3" style="width:65.5%"><span>65.5</span></div></div>
          </div>
        </div>
        
        <div class="bar-group">
          <div class="bar-label">C-Eval</div>
          <div class="bar-tracks">
            <div class="bar-track"><div class="bar-fill v4pro" style="width:93.1%"><span>93.1</span></div></div>
            <div class="bar-track"><div class="bar-fill v4flash" style="width:92.1%"><span>92.1</span></div></div>
            <div class="bar-track"><div class="bar-fill v3" style="width:90.4%"><span>90.4</span></div></div>
          </div>
        </div>
        
        <div class="bar-group">
          <div class="bar-label">HumanEval</div>
          <div class="bar-tracks">
            <div class="bar-track"><div class="bar-fill v4pro" style="width:76.8%"><span>76.8</span></div></div>
            <div class="bar-track"><div class="bar-fill v4flash" style="width:69.5%"><span>69.5</span></div></div>
            <div class="bar-track"><div class="bar-fill v3" style="width:62.8%"><span>62.8</span></div></div>
          </div>
        </div>
        
        <div class="bar-group">
          <div class="bar-label">MATH</div>
          <div class="bar-tracks">
            <div class="bar-track"><div class="bar-fill v4pro" style="width:64.5%"><span>64.5</span></div></div>
            <div class="bar-track"><div class="bar-fill v4flash" style="width:57.4%"><span>57.4</span></div></div>
            <div class="bar-track"><div class="bar-fill v3" style="width:60.5%"><span>60.5</span></div></div>
          </div>
        </div>
        
        <div class="bar-group">
          <div class="bar-label">SuperGPQA</div>
          <div class="bar-tracks">
            <div class="bar-track"><div class="bar-fill v4pro" style="width:53.9%"><span>53.9</span></div></div>
            <div class="bar-track"><div class="bar-fill v4flash" style="width:46.5%"><span>46.5</span></div></div>
            <div class="bar-track"><div class="bar-fill v3" style="width:45.0%"><span>45.0</span></div></div>
          </div>
        </div>
        
        <div class="bar-group">
          <div class="bar-label">LongBench-V2</div>
          <div class="bar-tracks">
            <div class="bar-track"><div class="bar-fill v4pro" style="width:51.5%"><span>51.5</span></div></div>
            <div class="bar-track"><div class="bar-fill v4flash" style="width:44.7%"><span>44.7</span></div></div>
            <div class="bar-track"><div class="bar-fill v3" style="width:40.2%"><span>40.2</span></div></div>
          </div>
        </div>
      </div>
    </div>
  </div>
</section>

<!-- ====== FRONTIER COMPARISON ====== -->
<div class="section-divider"></div>
<section>
  <div class="container">
    <div class="reveal">
      <div class="section-label">Frontier Battle</div>
      <h2 class="section-title">V4-Pro-Max vs 前沿闭源模型</h2>
      <p class="section-desc">
        DeepSeek-V4-Pro-Max 在编程和推理任务上多项指标领先顶级闭源模型，确立开源最强地位。
      </p>
    </div>
    
    <div class="chart-section reveal" style="overflow-x:auto;">
      <table class="comparison-table">
        <thead>
          <tr>
            <th style="min-width:180px">基准测试</th>
            <th>Opus-4.6</th>
            <th>GPT-5.4</th>
            <th>Gemini-3.1</th>
            <th>V4-Pro Max</th>
          </tr>
        </thead>
        <tbody>
          <tr>
            <td>MMLU-Pro</td>
            <td>89.1</td>
            <td>87.5</td>
            <td class="best-cell">91.0</td>
            <td class="highlight-cell">87.5</td>
          </tr>
          <tr>
            <td>GPQA Diamond</td>
            <td>91.3</td>
            <td>93.0</td>
            <td class="best-cell">94.3</td>
            <td class="highlight-cell">90.1</td>
          </tr>
          <tr>
            <td>LiveCodeBench</td>
            <td>88.8</td>
            <td>—</td>
            <td>91.7</td>
            <td class="best-cell">93.5 🏆</td>
          </tr>
          <tr>
            <td>Codeforces Rating</td>
            <td>—</td>
            <td>3168</td>
            <td>3052</td>
            <td class="best-cell">3206 🏆</td>
          </tr>
          <tr>
            <td>HMMT 2026</td>
            <td>96.2</td>
            <td class="best-cell">97.7</td>
            <td>94.7</td>
            <td class="highlight-cell">95.2</td>
          </tr>
          <tr>
            <td>IMOAnswerBench</td>
            <td>75.3</td>
            <td class="best-cell">91.4</td>
            <td>81.0</td>
            <td class="highlight-cell">89.8</td>
          </tr>
          <tr>
            <td>Apex Shortlist</td>
            <td>85.9</td>
            <td>78.1</td>
            <td>89.1</td>
            <td class="best-cell">90.2 🏆</td>
          </tr>
          <tr>
            <td>SWE Verified</td>
            <td class="best-cell">80.8</td>
            <td>—</td>
            <td>80.6</td>
            <td class="highlight-cell">80.6</td>
          </tr>
          <tr>
            <td>MCPAtlas Public</td>
            <td class="best-cell">73.8</td>
            <td>67.2</td>
            <td>69.2</td>
            <td class="highlight-cell">73.6</td>
          </tr>
          <tr>
            <td>BrowseComp</td>
            <td>83.7</td>
            <td>82.7</td>
            <td class="best-cell">85.9</td>
            <td class="highlight-cell">83.4</td>
          </tr>
        </tbody>
      </table>
    </div>
  </div>
</section>

<!-- ====== REASONING MODES ====== -->
<div class="section-divider"></div>
<section>
  <div class="container">
    <div class="reveal">
      <div class="section-label">Reasoning Modes</div>
      <h2 class="section-title">三种推理模式，灵活适配</h2>
      <p class="section-desc">
        V4 系列支持三种推理努力度模式，从快速直觉到深度思考，让用户根据场景灵活选择。
      </p>
    </div>
    
    <div class="modes-grid reveal">
      <div class="mode-card">
        <div class="mode-icon">⚡</div>
        <span class="mode-tag">Non-Think</span>
        <h4>直觉响应</h4>
        <p>快速、直觉性的回答，适合日常简单任务和低风险决策场景。无需深度思考链，极速响应。</p>
      </div>
      <div class="mode-card">
        <div class="mode-icon">🧠</div>
        <span class="mode-tag">Think High</span>
        <h4>深度分析</h4>
        <p>有意识的逻辑推理分析，通过 &lt;think&gt; 思考链实现更精准的复杂问题求解与规划。</p>
      </div>
      <div class="mode-card">
        <div class="mode-icon">🔥</div>
        <span class="mode-tag">Think Max</span>
        <h4>极限推理</h4>
        <p>将推理能力推向极限，探索模型推理边界。建议至少 384K 上下文窗口，适合最具挑战性的问题。</p>
      </div>
    </div>
  </div>
</section>

<!-- ====== TRAINING ====== -->
<div class="section-divider"></div>
<section>
  <div class="container">
    <div class="reveal">
      <div class="section-label">Training</div>
      <h2 class="section-title">训练与后训练流程</h2>
      <p class="section-desc">
        在超过 32 万亿高质量 Token 上预训练，配合创新的两阶段后训练范式，打造顶级开源模型。
      </p>
    </div>
    
    <div class="training-stats reveal">
      <div class="train-stat">
        <div class="num">32T+</div>
        <div class="desc">预训练 Tokens</div>
      </div>
      <div class="train-stat">
        <div class="num">FP8</div>
        <div class="desc">混合精度训练</div>
      </div>
      <div class="train-stat">
        <div class="num">GRPO</div>
        <div class="desc">强化学习策略</div>
      </div>
      <div class="train-stat">
        <div class="num">MIT</div>
        <div class="desc">开源许可证</div>
      </div>
    </div>
    
    <div class="chart-section reveal" style="margin-top:40px;">
      <h3 style="font-family:var(--font-heading);font-size:1.1rem;margin-bottom:24px;">后训练两阶段范式</h3>
      <div style="display:grid;grid-template-columns:1fr auto 1fr;gap:24px;align-items:center;">
        <div style="padding:24px;background:rgba(79,140,255,0.06);border-radius:var(--radius-md);border:1px solid rgba(79,140,255,0.15);">
          <div style="font-family:var(--font-mono);font-size:0.7rem;color:var(--accent-blue);margin-bottom:8px;">STAGE 1</div>
          <h4 style="font-family:var(--font-heading);font-size:1rem;margin-bottom:8px;">领域专家独立培养</h4>
          <p style="font-size:0.82rem;color:var(--text-secondary);line-height:1.7;">通过 SFT 和 GRPO 强化学习，在各个专业领域（编程、数学、推理、知识）独立培养领域专家。</p>
        </div>
        <div style="font-size:2rem;color:var(--accent-blue);">→</div>
        <div style="padding:24px;background:rgba(139,92,246,0.06);border-radius:var(--radius-md);border:1px solid rgba(139,92,246,0.15);">
          <div style="font-family:var(--font-mono);font-size:0.7rem;color:var(--accent-purple);margin-bottom:8px;">STAGE 2</div>
          <h4 style="font-family:var(--font-heading);font-size:1rem;margin-bottom:8px;">统一模型整合蒸馏</h4>
          <p style="font-size:0.82rem;color:var(--text-secondary);line-height:1.7;">通过 on-policy 蒸馏，将各领域专家的独特能力统一整合到单一模型中，实现全面能力融合。</p>
        </div>
      </div>
    </div>
  </div>
</section>

<!-- ====== EVOLUTION TIMELINE ====== -->
<div class="section-divider"></div>
<section>
  <div class="container">
    <div class="reveal">
      <div class="section-label">Evolution</div>
      <h2 class="section-title">DeepSeek 模型演进之路</h2>
      <p class="section-desc">
        从 V2 到 V4，DeepSeek 以惊人的速度不断突破大模型的能力边界。
      </p>
    </div>
    
    <div class="timeline reveal">
      <div class="timeline-item">
        <div class="year">2024 年中</div>
        <h4>DeepSeek-V2</h4>
        <p>首次引入 MLA（多头潜在注意力）和 DeepSeekMoE 架构，奠定高效 MoE 基础。236B 参数，21B 激活。</p>
      </div>
      <div class="timeline-item">
        <div class="year">2024 年末</div>
        <h4>DeepSeek-V3</h4>
        <p>671B 参数/37B 激活的 MoE 巨作，首创无辅助损失负载均衡和多 Token 预测。仅用 2.788M H800 GPU 小时完成训练。</p>
      </div>
      <div class="timeline-item">
        <div class="year">2025 年初</div>
        <h4>DeepSeek-R1 / V3.1 / V3.2</h4>
        <p>R1 系列引入长链思维推理，V3.1 和 V3.2 持续迭代优化。V3.2-Speciale 展示特化能力。</p>
      </div>
      <div class="timeline-item">
        <div class="year">2026 年 4 月</div>
        <h4>DeepSeek-V4 🚀</h4>
        <p>架构全面革新，CSA+HCA 混合注意力 + mHC 超连接 + Muon 优化器。1.6T 旗舰 Pro + 284B 高效 Flash，原生百万 Token 上下文。</p>
      </div>
    </div>
  </div>
</section>

<!-- ====== FOOTER ====== -->
<footer class="footer">
  <div class="container">
    <p style="font-family:var(--font-heading);font-size:1.2rem;font-weight:600;margin-bottom:8px;">
      DeepSeek V4 — Open Source, Open Future.
    </p>
    <p>MIT 许可证开源 · 模型权重和代码均可自由使用</p>
    <div class="links">
      <a href="https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" target="_blank">🤗 HuggingFace</a>
      <a href="https://chat.deepseek.com/" target="_blank">💬 在线体验</a>
      <a href="https://github.com/deepseek-ai" target="_blank">⭐ GitHub</a>
      <a href="https://www.deepseek.com/" target="_blank">🏠 官网</a>
    </div>
    <p style="margin-top:32px;font-size:0.75rem;color:rgba(255,255,255,0.3);">
      本文数据来源于 DeepSeek 官方技术报告与 HuggingFace 模型页面 · 2026.04.24
    </p>
  </div>
</footer>

<script>
// Scroll reveal animation
const observer = new IntersectionObserver((entries) => {
  entries.forEach(entry => {
    if (entry.isIntersecting) {
      entry.target.classList.add('visible');
    }
  });
}, {
  threshold: 0.1,
  rootMargin: '0px 0px -60px 0px'
});

document.querySelectorAll('.reveal').forEach(el => observer.observe(el));

// Animate bar chart on scroll
const barObserver = new IntersectionObserver((entries) => {
  entries.forEach(entry => {
    if (entry.isIntersecting) {
      entry.target.querySelectorAll('.bar-fill').forEach((bar, i) => {
        setTimeout(() => {
          bar.style.width = bar.style.width;
        }, i * 80);
      });
    }
  });
}, { threshold: 0.3 });

document.querySelectorAll('.bar-chart').forEach(el => barObserver.observe(el));
</script>

</body>
</html>