-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
1866 lines (1749 loc) · 129 KB
/
index.html
File metadata and controls
1866 lines (1749 loc) · 129 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<html lang="en" class="scroll-smooth">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>SGI-Bench — Scientific General Intelligence</title>
<meta name="description" content="SGI-Bench: Scientist-aligned benchmark for Scientific General Intelligence across Deliberation, Conception, Action, and Perception.">
<link rel="icon" href="data:image/svg+xml,<svg xmlns=%22http://www.w3.org/2000/svg%22 viewBox=%220 0 100 100%22><text y=%22.9em%22 font-size=%2290%22>🔬</text></svg>">
<script src="https://cdn.tailwindcss.com"></script>
<link href="https://fonts.googleapis.com/css2?family=Inter:wght@300;400;500;600;700&display=swap" rel="stylesheet">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.4.0/css/all.min.css">
<script>
// 说明：Tailwind 配置仅用于样式定制（字体、颜色、动画等），不会改变页面逻辑或文本内容。
// 动画滚动速度：下方 'scroll' 与 'scroll-reverse' 的持续时间为 60s（越小越快，越大越慢）。
tailwind.config = {
darkMode: ['class', '[data-theme="dark"]'],
theme: {
extend: {
fontFamily: { sans: ['Inter', 'sans-serif'] },
colors: {
brand: { 50: '#f8fafc', 900: '#0f172a' },
sgi: '#6366f1',
seevo: '#06b6d4',
sfe: '#10b981',
claw: '#f59e0b',
mark: '#ec4899'
},
animation: {
'scroll': 'scroll 60s linear infinite',
'scroll-reverse': 'scrollReverse 60s linear infinite',
},
keyframes: {
scroll: {
'0%': { transform: 'translateX(0)' },
'100%': { transform: 'translateX(-50%)' },
},
scrollReverse: {
'0%': { transform: 'translateX(-50%)' },
'100%': { transform: 'translateX(0)' },
}
}
}
}
}
</script>
<style>
/* ================================================================================================
SGIWorld Theme System - Light/Dark Mode Support
================================================================================================ */
:root {
--img-h-desktop: 220px;
--img-w-desktop: 330px;
--img-h-mobile: 140px;
--img-w-mobile: 210px;
/* SGIWorld Light Theme */
--color-bg: #f8fafc;
--color-bg-alt: #ffffff;
--color-text: #1e293b;
--color-text-muted: #64748b;
--color-border: rgba(0, 0, 0, 0.08);
--glass-bg: rgba(255, 255, 255, 0.75);
--glass-border: rgba(255, 255, 255, 0.8);
--dot-color: #e2e8f0;
--color-sgi: #6366f1;
}
[data-theme="dark"] {
--color-bg: #0B0F19;
--color-bg-alt: #111827;
--color-text: #f1f5f9;
--color-text-muted: #94a3b8;
--color-border: rgba(255, 255, 255, 0.1);
--glass-bg: rgba(17, 24, 39, 0.75);
--glass-border: rgba(55, 65, 81, 0.5);
--dot-color: #1e293b;
}
body {
background-color: var(--color-bg);
background-image:
radial-gradient(var(--dot-color) 1px, transparent 1px),
radial-gradient(var(--dot-color) 1px, transparent 1px);
background-size: 40px 40px;
background-position: 0 0, 20px 20px;
color: var(--color-text);
overflow-x: hidden;
transition: background-color 0.3s, color 0.3s;
}
.glass {
background: var(--glass-bg);
backdrop-filter: blur(12px);
-webkit-backdrop-filter: blur(12px);
border: 1px solid var(--glass-border);
box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.05);
}
[data-theme="dark"] .glass {
box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.3);
}
/* Dark mode text color overrides */
[data-theme="dark"] .text-slate-900 { color: #f1f5f9 !important; }
[data-theme="dark"] .text-slate-800 { color: #e2e8f0 !important; }
[data-theme="dark"] .text-slate-700 { color: #cbd5e1 !important; }
[data-theme="dark"] .text-slate-600 { color: #94a3b8 !important; }
[data-theme="dark"] .text-slate-500 { color: #64748b !important; }
[data-theme="dark"] .bg-white { background-color: var(--color-bg-alt) !important; }
[data-theme="dark"] .bg-slate-50 { background-color: rgba(30, 41, 59, 0.5) !important; }
[data-theme="dark"] .bg-slate-100 { background-color: rgba(30, 41, 59, 0.8) !important; }
[data-theme="dark"] .border-slate-200 { border-color: var(--color-border) !important; }
[data-theme="dark"] .border-slate-100 { border-color: var(--color-border) !important; }
[data-theme="dark"] .divide-slate-100 > * + * { border-color: var(--color-border) !important; }
[data-theme="dark"] .bg-indigo-50 { background-color: rgba(99, 102, 241, 0.2) !important; }
[data-theme="dark"] .hover\:bg-slate-50:hover { background-color: rgba(30, 41, 59, 0.5) !important; }
[data-theme="dark"] .bg-white\/95 { background-color: rgba(17, 24, 39, 0.95) !important; }
[data-theme="dark"] th.cursor-pointer:hover { background-color: rgba(255, 255, 255, 0.05) !important; }
[data-theme="dark"] th i { color: #475569; }
[data-theme="dark"] th:hover i { color: #94a3b8; }
[data-theme="dark"] th.asc i.fa-sort-up { color: #f1f5f9; }
[data-theme="dark"] th.desc i.fa-sort-down { color: #f1f5f9; }
.fade-in-up {
animation: fadeInUp 0.8s cubic-bezier(0.16, 1, 0.3, 1) forwards;
opacity: 0;
transform: translateY(20px);
}
.delay-100 { animation-delay: 0.1s; }
.delay-200 { animation-delay: 0.2s; }
.delay-300 { animation-delay: 0.3s; }
@keyframes fadeInUp {
to { opacity: 1; transform: translateY(0); }
}
.group:hover .animate-scroll, .group:hover .animate-scroll-reverse {
animation-play-state: paused;
}
th.cursor-pointer:hover { background-color: rgba(0,0,0,0.02); }
th i { color: #cbd5e1; transition: color 0.2s; margin-left: 6px; }
th:hover i { color: #64748b; }
th.asc i.fa-sort-up { color: #0f172a; }
th.desc i.fa-sort-down { color: #0f172a; }
#lightbox { transition: opacity 0.3s ease; }
#lightbox.hidden { opacity: 0; pointer-events: none; }
#lightbox:not(.hidden) { opacity: 1; pointer-events: auto; }
.gallery-item {
background: var(--glass-bg);
backdrop-filter: blur(12px);
-webkit-backdrop-filter: blur(12px);
border: 1px solid var(--glass-border);
box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.05);
/* 间距与圆角 */
padding: 0.5rem; /* p-2 */
border-radius: 0.75rem; /* rounded-xl */
/* 布局与交互 */
flex-shrink: 0; /* flex-shrink-0 */
transition: transform 150ms ease; /* transition-transform */
cursor: pointer; /* cursor-pointer */
position: relative; /* relative */
overflow: hidden; /* overflow-hidden */
/* 尺寸由 CSS 变量控制（移动端） */
height: var(--img-h-mobile);
width: var(--img-w-mobile);
}
.gallery-item:hover {
transform: scale(1.02); /* hover:scale-[1.02] */
}
@media (min-width: 768px) {
.gallery-item {
height: var(--img-h-desktop);
width: var(--img-w-desktop);
}
}
.gallery-item img {
/* 展开 @apply 为原生 CSS，确保在无构建步骤时也能生效 */
width: 100%; /* w-full */
height: 100%; /* h-full */
object-fit: cover; /* object-cover */
border-radius: 0.5rem; /* rounded-lg */
}
/* Leaderboard 头图：白框与图片宽度联动且水平居中 */
:root {
--lb-img-w-desktop: 900px; /* 桌面端统一宽度（可调） */
--lb-img-w-mobile: 320px; /* 移动端统一宽度（可调） */
}
.leaderboard-figure-wrap {
display: flex; /* 使用块级 flex，便于 margin: auto 居中 */
align-items: center;
justify-content: center;
width: var(--lb-img-w-mobile); /* 宽度由变量控制，等比例缩放 */
margin-left: auto; /* 水平居中 */
margin-right: auto; /* 水平居中 */
max-width: 100%; /* 防止溢出父容器 */
}
.leaderboard-figure {
width: 100%; /* 固定宽度，保持等比例 */
height: auto; /* 高度自适应，不做限制 */
max-width: 100%;
object-fit: contain;
display: block;
margin-left: auto; /* 确保图片在白框内部水平居中 */
margin-right: auto; /* 与上同理 */
}
@media (min-width: 768px) {
.leaderboard-figure-wrap { width: var(--lb-img-w-desktop); }
}
</style>
</head>
<body class="antialiased">
<!-- 全局调参指南：
1) 尺寸（宽/高）：
- Gallery 卡片：CSS :root 变量 --img-h-desktop / --img-w-desktop / --img-h-mobile / --img-w-mobile（单位 px）
- Leaderboard 头图白框与宽度：:root 变量 --lb-img-w-desktop / --lb-img-w-mobile（单位 px）
- 灯箱预览：#lightbox-img 的 max-h-[90vh] 控制最大高度（相对视口高度）；需要更大/更小可改 90vh。
- 各组件内的尺寸类：如 w-8 h-8（图标）、h-16（导航栏高度）、min-h-[85vh]（Hero 高度），可按需调整。
2) 速度与等待：
- 横向滚动速度：Tailwind 动画 'scroll' / 'scroll-reverse' 的持续时间 60s（在 <head> 中 tailwind.config.animation）。
- 入场动画速度：.fade-in-up 动画时长 0.8s；延迟类 .delay-100/200/300 分别为 0.1/0.2/0.3s。
- 灯箱淡入淡出：#lightbox 的过渡时间 0.3s；点击关闭时脚本中的 setTimeout 200ms。
- Demo 播放速度：STREAM_SPEED（每字符毫秒，默认 25，越小越快）；对话气泡宽度：DIALOG_MAX_WIDTH（默认 700px）。
- Demo 等待节奏：脚本中的 setDemoTimeout/setDemoInterval 数值，例如 1000ms（阶段间隔）、180ms（进度条更新间隔）、2000ms（插入报告延迟）。
3) 其它：
- Gallery 每张卡片的间距：HTML 中 gallery-item 外层使用 mx-4，可调节横向间隙。
- 渐变遮罩宽度：左右遮罩 w-12（移动端）/ md:w-32（桌面端），可按需修改以改变边缘淡入区域。
以上改动均不影响页面逻辑，仅改变视觉与节奏。 -->
<!-- ==================================================================================================== -->
<!-- 导航栏（Navbar）板块：页面顶部导航与图标。此处仅用于跳转与展示，不涉及交互逻辑改动。 -->
<!-- ==================================================================================================== -->
<!-- Navbar -->
<nav class="fixed w-full z-50 glass border-b border-white/50 transition-all duration-300">
<div class="max-w-7xl mx-auto px-4 sm:px-6 lg:px-8">
<!-- 导航栏高度由 h-16 控制（约 64px）；可改为 h-20 等以增大高度 -->
<div class="flex justify-between items-center h-16">
<div class="hidden md:flex space-x-8">
<a href="#about" class="text-sm font-medium text-slate-600 hover:text-slate-900 transition-colors">About</a>
<a href="#agent-demo" class="text-sm font-medium text-slate-600 hover:text-slate-900 transition-colors">Demo</a>
<a href="#gallery" class="text-sm font-medium text-slate-600 hover:text-slate-900 transition-colors">Gallery</a>
<a href="#leaderboard" class="text-sm font-medium text-slate-600 hover:text-slate-900 transition-colors">Leaderboard</a>
</div>
<div class="md:hidden flex items-center">
<button onclick="document.getElementById('mobile-menu').classList.toggle('hidden')" class="text-slate-500 hover:text-slate-900 focus:outline-none">
<i class="fas fa-bars text-xl"></i>
</button>
</div>
<div class="hidden md:flex items-center gap-4">
<button onclick="toggleTheme()" class="p-2 rounded-lg hover:bg-slate-100 transition-colors" title="Toggle theme">
<i class="fas fa-moon text-slate-600 theme-icon-dark"></i>
<i class="fas fa-sun text-slate-600 theme-icon-light hidden"></i>
</button>
<a href="https://arxiv.org/pdf/2512.16969" target="_blank" rel="noopener noreferrer" class="hover:opacity-80 transition-opacity" title="arXiv">
<img src="images/arxiv.jpg" alt="arXiv" class="w-8 h-8 rounded-full border border-slate-200">
</a>
<a href="https://huggingface.co/collections/InternScience/sgi-bench" target="_blank" rel="noopener noreferrer" class="hover:opacity-80 transition-opacity" title="HuggingFace">
<img src="images/huggingface.jpg" alt="HuggingFace" class="w-8 h-8 rounded-full border border-slate-200">
</a>
<a href="https://github.com/InternScience/SGI-Bench" target="_blank" rel="noopener noreferrer" class="hover:opacity-80 transition-opacity" title="GitHub">
<img src="images/github.jpg" alt="GitHub" class="w-8 h-8 rounded-full border border-slate-200">
</a>
</div>
</div>
</div>
<div id="mobile-menu" class="hidden md:hidden bg-white/95 backdrop-blur-md border-b border-slate-200 absolute w-full">
<div class="px-4 pt-2 pb-4 space-y-1">
<a href="#about" class="block px-3 py-2 rounded-md text-base font-medium text-slate-700 hover:bg-slate-50">About</a>
<a href="#agent-demo" class="block px-3 py-2 rounded-md text-base font-medium text-slate-700 hover:bg-slate-50">Demo</a>
<a href="#gallery" class="block px-3 py-2 rounded-md text-base font-medium text-slate-700 hover:bg-slate-50">Gallery</a>
<a href="#leaderboard" class="block px-3 py-2 rounded-md text-base font-medium text-slate-700 hover:bg-slate-50">Leaderboard</a>
</div>
</div>
</nav>
<!-- ==================================================================================================== -->
<!-- 介绍（Hero）板块：封面标题、简介段落与预览图。仅视觉呈现，不改变任何逻辑与文本。 -->
<!-- 图像点击后会打开灯箱（Lightbox），此处不用修改功能。 -->
<!-- ==================================================================================================== -->
<!-- Hero -->
<!-- Hero 高度由 min-h-[85vh] 控制（占视口高度的 85%）；pt-32/pb-20 控制上下内边距 -->
<section id="home" class="relative pt-32 pb-20 px-4 max-w-7xl mx-auto min-h-[85vh] flex flex-col justify-center text-center">
<div class="max-w-4xl mx-auto space-y-8">
<div class="fade-in-up inline-flex items-center px-3 py-1 rounded-full bg-white border border-slate-200 text-slate-600 text-xs font-medium shadow-sm">
<span class="w-2 h-2 rounded-full bg-indigo-500 mr-2"></span> SGI-Bench v1.0 — Scientist-aligned Evaluation
</div>
<h1 class="fade-in-up delay-100 text-5xl md:text-7xl font-bold text-slate-900 leading-tight">
Scientific General <br>
<span class="text-transparent bg-clip-text bg-gradient-to-r from-indigo-500 to-purple-600">Intelligence</span>
</h1>
<p class="fade-in-up delay-200 text-xl text-slate-600 max-w-3xl mx-auto leading-relaxed px-4">
Scientific General Intelligence (SGI) is defined as an AI system that can autonomously navigate the full, iterative cycle of scientific inquiry—Deliberation, Conception, Action, and Perception—with the versatility and proficiency of a human scientist. SGI-Bench operationalizes this definition via four scientist-aligned task families: deep research, idea generation, dry/wet experiments, and multimodal experimental reasoning. The benchmark spans 10 disciplines and more than 1,000 expert-curated samples inspired by Science's 125 Big Questions.
</p>
<div class="fade-in-up delay-300 mt-12 p-2 glass rounded-2xl mx-auto max-w-full sm:max-w-4xl">
<img src="paper/imgs/teaser.png" alt="SGI Overview" class="w-full h-auto rounded-xl shadow-sm cursor-pointer hover:opacity-95 transition" onclick="openModal(this.src)" onerror="this.parentElement.innerHTML='<div class=\'h-64 flex items-center justify-center bg-slate-100 rounded-xl text-slate-400\'>Image teaser.jpg missing</div>'">
</div>
</div>
</section>
<!-- ==================================================================================================== -->
<!-- 关于（About）板块：框架、数据构建、评估框架、TTRL、以及 Demo 子板块入口。 -->
<!-- 此板块包含多个子区域，每个区域仅展示图文内容。 -->
<!-- ==================================================================================================== -->
<!-- About -->
<section id="about" class="py-24 px-4 max-w-7xl mx-auto space-y-32">
<!-- SGI & Tasks -->
<div class="grid md:grid-cols-2 gap-16 items-center">
<div class="fade-in-up order-2 md:order-1">
<div class="glass p-2 rounded-2xl hover:shadow-lg transition-shadow duration-300">
<img src="paper/imgs/pipeline.png" alt="SGI-Bench Pipeline" class="rounded-xl w-full cursor-pointer hover:opacity-95 transition" onclick="openModal(this.src)" onerror="this.src='https://placehold.co/800x600?text=Pipeline+Image'">
</div>
</div>
<div class="order-1 md:order-2 space-y-6 fade-in-up delay-100">
<h2 class="text-3xl font-bold text-slate-900">SGI Framework & Tasks</h2>
<div class="mt-4 grid sm:grid-cols-2 gap-4">
<div class="flex gap-4 p-4 glass rounded-xl border-l-4 border-blue-500">
<div class="flex-shrink-0 w-10 h-10 rounded-full bg-blue-50 flex items-center justify-center text-blue-600"><i class="fas fa-magnifying-glass"></i></div>
<div>
<h4 class="font-semibold text-slate-900">Deliberation</h4>
<p class="text-sm text-slate-500">Deep research with multi-hop retrieval and meta-analysis style quantitative synthesis.</p>
</div>
</div>
<div class="flex gap-4 p-4 glass rounded-xl border-l-4 border-purple-500">
<div class="flex-shrink-0 w-10 h-10 rounded-full bg-purple-50 flex items-center justify-center text-purple-600"><i class="fas fa-lightbulb"></i></div>
<div>
<h4 class="font-semibold text-slate-900">Conception</h4>
<p class="text-sm text-slate-500">Structured idea generation and multi-dimensional comparative evaluation.</p>
</div>
</div>
<div class="flex gap-4 p-4 glass rounded-xl border-l-4 border-green-500">
<div class="flex-shrink-0 w-10 h-10 rounded-full bg-green-50 flex items-center justify-center text-green-600"><i class="fas fa-code"></i></div>
<div>
<h4 class="font-semibold text-slate-900">Action</h4>
<p class="text-sm text-slate-500">Scientific experiments: dry (code generation) and wet (lab protocol development).</p>
</div>
</div>
<div class="flex gap-4 p-4 glass rounded-xl border-l-4 border-orange-500">
<div class="flex-shrink-0 w-10 h-10 rounded-full bg-orange-50 flex items-center justify-center text-orange-600"><i class="fas fa-eye"></i></div>
<div>
<h4 class="font-semibold text-slate-900">Perception</h4>
<p class="text-sm text-slate-500">Multimodal reasoning: process, observation, simulation, experiment, visualization images.</p>
</div>
</div>
</div>
<p class="text-base md:text-base text-slate-600 leading-relaxed">
Grounded in the Practical Inquiry Model (PIM), SGI‑Bench views science as an iterative cycle that links deliberation, conception, action and perception at a high level. Under this lens, Scientific General Intelligence (SGI) denotes an AI’s capacity to traverse that cycle coherently and autonomously—integrating knowledge retrieval, idea formation, action execution, and interpretation into a unified loop of inquiry.
</p>
</div>
</div>
<!-- Scientist-Aligned Data Construction -->
<div class="grid md:grid-cols-2 gap-16 items-center">
<div class="fade-in-up order-2 md:order-1">
<div class="glass p-2 rounded-2xl hover:shadow-lg transition-shadow duration-300">
<img src="paper/imgs/subjects.png" alt="Scientist-Aligned Data Construction" class="rounded-xl w-full cursor-pointer hover:opacity-95 transition" onclick="openModal(this.src)" onerror="this.src='https://placehold.co/800x600/f1f5f9/94a3b8?text=Data+Construction+Image'">
</div>
</div>
<div class="order-1 md:order-2 space-y-6 fade-in-up delay-100">
<h2 class="text-3xl font-bold text-slate-900">Scientist-Aligned Data</h2>
<div class="mt-4 grid sm:grid-cols-2 gap-4">
<div class="flex gap-4 p-4 glass rounded-xl border-l-4 border-teal-500">
<div class="flex-shrink-0 w-10 h-10 rounded-full bg-teal-50 flex items-center justify-center text-teal-600"><i class="fas fa-database"></i></div>
<div>
<h4 class="font-semibold text-slate-900">Raw Corpus</h4>
<p class="text-sm text-slate-500">Expert-curated texts/images across 10 domains from Science’s 125 Big Questions.</p>
</div>
</div>
<div class="flex gap-4 p-4 glass rounded-xl border-l-4 border-indigo-500">
<div class="flex-shrink-0 w-10 h-10 rounded-full bg-indigo-50 flex items-center justify-center text-indigo-600"><i class="fas fa-pen-nib"></i></div>
<div>
<h4 class="font-semibold text-slate-900">Question Construction</h4>
<p class="text-sm text-slate-500">100+ Master's and PhD holders; continuous expert review for scientific value.</p>
</div>
</div>
<div class="flex gap-4 p-4 glass rounded-xl border-l-4 border-amber-500">
<div class="flex-shrink-0 w-10 h-10 rounded-full bg-amber-50 flex items-center justify-center text-amber-600"><i class="fas fa-check-double"></i></div>
<div>
<h4 class="font-semibold text-slate-900">Data Cleaning</h4>
<p class="text-sm text-slate-500">Rule-based validation, model checks, expert QA for executability and unique answers.</p>
</div>
</div>
<div class="flex gap-4 p-4 glass rounded-xl border-l-4 border-rose-500">
<div class="flex-shrink-0 w-10 h-10 rounded-full bg-rose-50 flex items-center justify-center text-rose-600"><i class="fas fa-signal"></i></div>
<div>
<h4 class="font-semibold text-slate-900">Difficulty Filtering</h4>
<p class="text-sm text-slate-500">Remove samples solvable by >50% strong LLMs to maintain high challenge.</p>
</div>
</div>
</div>
<p class="text-base md:text-base text-slate-600 leading-relaxed">
SGI-Bench data is scientist-aligned and high-fidelity: an expert-sourced corpus spanning 10 disciplines (inspired by Science’s 125 Big Questions), questions constructed by 100+ Master's and PhD holders with continuous scientist-in-the-loop review, multi-stage cleaning (rules + model checks + expert QA) to ensure executability and unique answers, and difficulty filtering that removes items solved by >50% strong LLMs—yielding authentic, challenging, and broadly representative scientific tasks.
</p>
</div>
</div>
<!-- Agentic Evaluation Framework -->
<div class="grid md:grid-cols-2 gap-16 items-center">
<div class="order-1 md:order-2 space-y-6 fade-in-up delay-100">
<h2 class="text-3xl font-bold text-slate-900">Agentic Evaluation Framework</h2>
<div class="mt-4 grid sm:grid-cols-2 gap-4">
<div class="flex gap-4 p-4 glass rounded-xl border-l-4 border-emerald-500">
<div class="flex-shrink-0 w-10 h-10 rounded-full bg-emerald-50 flex items-center justify-center text-emerald-600"><i class="fas fa-list-check"></i></div>
<div>
<h4 class="font-semibold text-slate-900">Four Stages</h4>
<p class="text-sm text-slate-500">Question Selection → Metric Customization → Predict & Eval → Report Generation.</p>
</div>
</div>
<div class="flex gap-4 p-4 glass rounded-xl border-l-4 border-cyan-500">
<div class="flex-shrink-0 w-10 h-10 rounded-full bg-cyan-50 flex items-center justify-center text-cyan-600"><i class="fas fa-toolbox"></i></div>
<div>
<h4 class="font-semibold text-slate-900">Tool Pool</h4>
<p class="text-sm text-slate-500">Web search, PDF parser, Python interpreter, file reader, metric functions.</p>
</div>
</div>
<div class="flex gap-4 p-4 glass rounded-xl border-l-4 border-blue-500">
<div class="flex-shrink-0 w-10 h-10 rounded-full bg-blue-50 flex items-center justify-center text-blue-600"><i class="fas fa-gauge-high"></i></div>
<div>
<h4 class="font-semibold text-slate-900">Task Metrics</h4>
<p class="text-sm text-slate-500">EM/SLA; Implementation Similarity; PassAll@k/SER; MCA/RV.</p>
</div>
</div>
<div class="flex gap-4 p-4 glass rounded-xl border-l-4 border-orange-500">
<div class="flex-shrink-0 w-10 h-10 rounded-full bg-orange-50 flex items-center justify-center text-orange-600"><i class="fas fa-sliders"></i></div>
<div>
<h4 class="font-semibold text-slate-900">Customizable</h4>
<p class="text-sm text-slate-500">Add scientist-aligned metrics (e.g., rigor, feasibility) on demand.</p>
</div>
</div>
</div>
<p class="text-base md:text-base text-slate-600 leading-relaxed">
An agent-based evaluation stack coordinating specialized agents and tools to assess models end-to-end with task-specific and customizable metrics. By formalizing question selection, metric construction, scoring, and reporting into traceable stages, it strengthens reproducibility, mitigates evaluator–model coupling bias, and offers scientist‑aligned, actionable insights for model selection and iteration.
</p>
</div>
<div class="fade-in-up order-2 md:order-1">
<div class="glass p-2 rounded-2xl hover:shadow-lg transition-shadow duration-300">
<img src="paper/imgs/evaluation-framework.png" alt="Evaluation Framework" class="rounded-xl w-full cursor-pointer hover:opacity-95 transition" onclick="openModal(this.src)" onerror="this.src='https://placehold.co/800x600?text=Evaluation+Framework'">
</div>
</div>
</div>
<!-- Test-Time Reinforcement Learning (TTRL) -->
<div class="grid md:grid-cols-2 gap-16 items-center">
<div class="fade-in-up order-2 md:order-1">
<div class="glass p-2 rounded-2xl hover:shadow-lg transition-shadow duration-300">
<img src="paper/imgs/grpo_reward_curves.png" alt="TTRL Training Dynamics" class="rounded-xl w-full cursor-pointer hover:opacity-95 transition" onclick="openModal(this.src)" onerror="this.src='https://placehold.co/800x600?text=TTRL+Reward+Curves'">
</div>
</div>
<div class="order-1 md:order-2 space-y-6 fade-in-up delay-100">
<h2 class="text-3xl font-bold text-slate-900">Test-Time Reinforcement Learning (TTRL)</h2>
<div class="mt-4 grid sm:grid-cols-2 gap-4">
<div class="flex gap-4 p-4 glass rounded-xl border-l-4 border-emerald-500">
<div class="flex-shrink-0 w-10 h-10 rounded-full bg-emerald-50 flex items-center justify-center text-emerald-600"><i class="fas fa-bullseye"></i></div>
<div>
<h4 class="font-semibold text-slate-900">Objective</h4>
<p class="text-sm text-slate-500">Handle no-ground-truth idea generation by optimizing novelty at test time with online retrieval as the moving baseline.</p>
</div>
</div>
<div class="flex gap-4 p-4 glass rounded-xl border-l-4 border-indigo-500">
<div class="flex-shrink-0 w-10 h-10 rounded-full bg-indigo-50 flex items-center justify-center text-indigo-600"><i class="fas fa-award"></i></div>
<div>
<h4 class="font-semibold text-slate-900">Reward Design</h4>
<p class="text-sm text-slate-500">R = R_format + R_novelty. Enforce XML format (<think>, <answer>); reward embedding dissimilarity from retrieved works using a gating threshold.</p>
</div>
</div>
<div class="flex gap-4 p-4 glass rounded-xl border-l-4 border-sky-500">
<div class="flex-shrink-0 w-10 h-10 rounded-full bg-sky-50 flex items-center justify-center text-sky-600"><i class="fas fa-gears"></i></div>
<div>
<h4 class="font-semibold text-slate-900">Training Setup</h4>
<p class="text-sm text-slate-500">GRPO on Qwen3‑8B (ms-swift); group sampling G=8, high temperature, bfloat16; online retrieval n=4.</p>
</div>
</div>
<div class="flex gap-4 p-4 glass rounded-xl border-l-4 border-amber-500">
<div class="flex-shrink-0 w-10 h-10 rounded-full bg-amber-50 flex items-center justify-center text-amber-600"><i class="fas fa-chart-line"></i></div>
<div>
<h4 class="font-semibold text-slate-900">Dynamics & Results</h4>
<p class="text-sm text-slate-500">Format reward saturates quickly; novelty steadily increases. Average novelty improves from 49.36 → 62.06 without labels.</p>
</div>
</div>
</div>
<p class="text-base md:text-base text-slate-600 leading-relaxed">
TTRL converts open-ended scientific exploration into a measurable test-time optimization process and can be extended to multi-objective scientist-aligned rewards (rigor, feasibility, safety, cost). In practice, it improves idea novelty without labels by coupling strict output structure with retrieval-grounded rewards, and generalizes to multi‑objective optimization that balances creativity with rigor and feasibility—making scientific ideation auditable and adaptable at inference time.
</p>
</div>
</div>
<!-- ==================================================================================================== -->
<!-- Demo（SGIEvalAgent 演示）子板块：包含对话流演示与可调常量（宽度与播放速度）。 -->
<!-- 重要：演示板块的全局常量请见下方脚本中的“SGIEvalAgent Demo 常量入口”注释行后，便于快速定位修改。 -->
<!-- ==================================================================================================== -->
<!-- SGIEvalAgent Chat Demo -->
<div class="space-y-1">
<h2 id="agent-demo" class="text-3xl font-bold text-slate-900 mb-4 scroll-mt-24 md:scroll-mt-28">SGIEvalAgent Demo</h2>
<div class="fade-in-up mt-2">
<div class="glass rounded-2xl p-4 md:p-6">
<div class="flex items-center justify-end mb-4">
<button id="redo-demo" class="text-xs px-2.5 py-1 rounded-md bg-slate-900 text-white hover:opacity-90">Replay</button>
</div>
<div id="agent-chat" class="mt-4 space-y-3 pr-1"></div>
</div>
</div>
</div>
</section>
<!-- ==================================================================================================== -->
<!-- 图库（Gallery）板块：横向滚动图片展示。滚动速度由 Tailwind 动画持续时间控制（见 head 中配置）。 -->
<!-- 鼠标悬停时暂停滚动，仅视觉效果。图片点击打开灯箱。 -->
<!-- ==================================================================================================== -->
<!-- Gallery -->
<section id="gallery" class="py-20 border-y border-slate-200 bg-white/40 backdrop-blur-sm overflow-hidden">
<div class="max-w-7xl mx-auto px-4 mb-12 text-center fade-in-up">
<h2 class="text-3xl font-bold text-slate-900">Visual Analysis Cases</h2>
<p class="text-slate-500 mt-3 max-w-2xl mx-auto">
Explore scientific multimodal inputs across process, observation, experiment, simulation, and visualization images.
<span class="hidden sm:inline text-slate-400 text-sm">(Click images to expand)</span>
</p>
</div>
<div class="relative w-full fade-in-up delay-100 space-y-6">
<!-- 左侧遮罩宽度：移动端 w-12，桌面端 md:w-32；调大可增强边缘淡入区域 -->
<div class="absolute left-0 top-0 bottom-0 w-12 md:w-32 z-10 bg-gradient-to-r from-[#f8fafc] to-transparent pointer-events-none"></div>
<!-- 右侧遮罩宽度：移动端 w-12，桌面端 md:w-32；与左侧保持一致 -->
<div class="absolute right-0 top-0 bottom-0 w-12 md:w-32 z-10 bg-gradient-to-l from-[#f8fafc] to-transparent pointer-events-none"></div>
<!-- Row 1 -->
<!-- Gallery 每行：卡片间距由子元素上的 mx-4 控制；横向滚动速度见 head 中 Tailwind 动画 60s -->
<div class="group flex overflow-hidden w-full">
<div class="flex w-max animate-scroll group-hover:[animation-play-state:paused]">
<div class="gallery-item mx-4"><img src="images/science_1.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_5.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_9.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_13.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_17.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_21.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_25.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_29.jpg" onclick="openModal(this.src)"></div>
<!-- Duplicate -->
<div class="gallery-item mx-4"><img src="images/science_1.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_5.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_9.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_13.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_17.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_21.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_25.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_29.jpg" onclick="openModal(this.src)"></div>
</div>
</div>
<!-- Row 2 -->
<div class="group flex overflow-hidden w-full">
<div class="flex w-max animate-scroll-reverse group-hover:[animation-play-state:paused]">
<div class="gallery-item mx-4"><img src="images/science_2.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_6.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_10.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_14.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_18.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_22.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_26.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_30.jpg" onclick="openModal(this.src)"></div>
<!-- Duplicate -->
<div class="gallery-item mx-4"><img src="images/science_2.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_6.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_10.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_14.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_18.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_22.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_26.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_30.jpg" onclick="openModal(this.src)"></div>
</div>
</div>
<!-- Row 3 -->
<div class="group flex overflow-hidden w-full">
<div class="flex w-max animate-scroll group-hover:[animation-play-state:paused]">
<div class="gallery-item mx-4"><img src="images/science_3.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_7.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_11.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_15.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_19.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_23.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_27.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_31.jpg" onclick="openModal(this.src)"></div>
<!-- Duplicate -->
<div class="gallery-item mx-4"><img src="images/science_3.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_7.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_11.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_15.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_19.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_23.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_27.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_31.jpg" onclick="openModal(this.src)"></div>
</div>
</div>
<!-- Row 4 -->
<div class="group flex overflow-hidden w-full">
<div class="flex w-max animate-scroll-reverse group-hover:[animation-play-state:paused]">
<div class="gallery-item mx-4"><img src="images/science_4.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_8.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_12.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_16.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_20.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_24.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_28.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_32.jpg" onclick="openModal(this.src)"></div>
<!-- Duplicate -->
<div class="gallery-item mx-4"><img src="images/science_4.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_8.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_12.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_16.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_20.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_24.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_28.jpg" onclick="openModal(this.src)"></div>
<div class="gallery-item mx-4"><img src="images/science_32.jpg" onclick="openModal(this.src)"></div>
</div>
</div>
</div>
</section>
<!-- ==================================================================================================== -->
<!-- 排行（Leaderboard）板块：两张表格（Idea Generation / Dry Experiment）。表格数据由脚本渲染。 -->
<!-- 修改提示：只需改脚本数组 IDEA_DATA / DRY_DATA 即可更新表格，不建议改表格结构。 -->
<!-- ==================================================================================================== -->
<!-- Leaderboard -->
<section id="leaderboard" class="py-8 px-4 max-w-7xl mx-auto scroll-mt-8 md:scroll-mt-12">
<div class="text-center mb-12 fade-in-up">
<h2 class="text-3xl font-bold text-slate-900">Leaderboard</h2>
</div>
<!-- Table - Overview (硬编码，交集模型；结构与其他表一致) -->
<div class="mb-24 fade-in-up">
<div class="flex flex-col sm:flex-row justify-between items-end mb-4 gap-4">
<h3 class="text-2xl font-bold text-slate-800 flex items-center gap-2">
<span class="w-8 h-8 rounded-lg bg-slate-100 text-slate-600 flex items-center justify-center text-sm"><i class="fas fa-table"></i></span>
Overview
</h3>
</div>
<div class="flex justify-center mb-3">
<div class="glass p-1 rounded-lg flex gap-2 text-sm w-fit overflow-x-auto">
<button onclick="filterTable('t-ov', 'all')" class="px-4 py-1.5 rounded-md bg-slate-900 text-white shadow transition-all filter-btn active whitespace-nowrap" data-t="t-ov" data-f="all">All Models</button>
<button onclick="filterTable('t-ov', 'open')" class="px-4 py-1.5 rounded-md text-slate-500 hover:text-slate-900 transition-all filter-btn whitespace-nowrap" data-t="t-ov" data-f="open">Open Source</button>
<button onclick="filterTable('t-ov', 'closed')" class="px-4 py-1.5 rounded-md text-slate-500 hover:text-slate-900 transition-all filter-btn whitespace-nowrap" data-t="t-ov" data-f="closed">Closed Source</button>
</div>
</div>
<div class="glass rounded-xl overflow-hidden ring-1 ring-slate-200 shadow-sm">
<div class="overflow-x-auto">
<table id="t-ov" class="w-full text-left text-sm whitespace-nowrap">
<thead>
<tr class="bg-slate-50/80 border-b border-slate-200 text-slate-500 font-semibold">
<th class="p-4 pl-6 cursor-pointer" onclick="sort('t-ov', 0)">Model <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t-ov', 1)">Type <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t-ov', 2)">Deep Research<i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t-ov', 3)">Idea Generation<i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t-ov', 4)">Dry Experiment<i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t-ov', 5)">Wet Experiment<i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t-ov', 6)">Experimental Reasoning<i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer bg-slate-100/50" onclick="sort('t-ov', 7)">SGI-Score <i class="fas fa-sort ml-1"></i></th>
</tr>
</thead>
<tbody class="divide-y divide-slate-100" id="t-ov-body"></tbody>
</table>
</div>
</div>
<div class="mt-4 text-slate-500 text-sm leading-relaxed bg-white/50 p-4 rounded-lg border border-slate-100">
<p><strong class="text-slate-700">Overview Results Across SGI-Bench Tasks:</strong> Aggregated performance across Deep Research, Idea Generation, Dry/Wet Experiment, and Experimental Reasoning. The scores for Deep Research are based on the exact match metric (the strictest metric). Idea Generation scores are the average of four metrics evaluating ideas. Dry Experiment scores are based on PassAll@5 (the strictest metric). Wet Experiment scores are the average of action sequence similarity and parameter accuracy. Experimental Reasoning scores are based on the multi-choice accuracy metric (the strictest metric). The SGI-Score is the average across these tasks, reflecting the overall capability of an AI model in various scientific research scenarios. Note: The Experimental Reasoning metrics for Qwen3-Max and Qwen3-8B come from their multimodal versions.</p>
</div>
</div>
<!-- Table 0: Deep Research Metrics -->
<div class="mb-24 fade-in-up delay-50">
<div class="flex flex-col sm:flex-row justify-between items-end mb-4 gap-4">
<h3 class="text-2xl font-bold text-slate-800 flex items-center gap-2">
<span class="w-8 h-8 rounded-lg bg-blue-100 text-blue-600 flex items-center justify-center text-sm"><i class="fas fa-magnifying-glass"></i></span>
Scientific Deep Research
</h3>
</div>
<div class="glass p-2 rounded-2xl hover:shadow-lg transition-shadow duration-300 mb-4 leaderboard-figure-wrap">
<img src="paper/imgs/deepresearch_case1.png" alt="Deep Research Case" class="rounded-xl cursor-pointer hover:opacity-95 transition leaderboard-figure" onclick="openModal(this.src)" onerror="this.src='https://placehold.co/1200x500?text=Deep+Research+Case'">
</div>
<div class="flex items-center justify-center relative mb-3">
<!-- Left: LLMs/Agents toggle -->
<div class="absolute left-0">
<div class="glass p-1 rounded-lg flex gap-2 text-sm w-fit overflow-x-auto">
<button onclick="setDeepGroup('llms')" class="px-4 py-1.5 rounded-md bg-slate-900 text-white shadow transition-all deep-btn active whitespace-nowrap" data-g="llms">LLMs</button>
<button onclick="setDeepGroup('agents')" class="px-4 py-1.5 rounded-md text-slate-500 hover:text-slate-900 transition-all deep-btn whitespace-nowrap" data-g="agents">Agents</button>
</div>
</div>
<!-- Center: Existing type filter (unchanged position) -->
<div class="glass p-1 rounded-lg flex gap-2 text-sm w-fit overflow-x-auto">
<button onclick="filterTable('t0', 'all')" class="px-4 py-1.5 rounded-md bg-slate-900 text-white shadow transition-all filter-btn active whitespace-nowrap" data-t="t0" data-f="all">All Models</button>
<button onclick="filterTable('t0', 'open')" class="px-4 py-1.5 rounded-md text-slate-500 hover:text-slate-900 transition-all filter-btn whitespace-nowrap" data-t="t0" data-f="open">Open Source</button>
<button onclick="filterTable('t0', 'closed')" class="px-4 py-1.5 rounded-md text-slate-500 hover:text-slate-900 transition-all filter-btn whitespace-nowrap" data-t="t0" data-f="closed">Closed Source</button>
</div>
</div>
<div class="glass rounded-xl overflow-hidden ring-1 ring-slate-200 shadow-sm">
<div class="overflow-x-auto">
<table id="t0" class="w-full text-left text-sm whitespace-nowrap">
<thead>
<tr class="bg-slate-50/80 border-b border-slate-200 text-slate-500 font-semibold">
<th class="p-4 pl-6 cursor-pointer" onclick="sort('t0', 0)">Model <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t0', 1)">Type <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer bg-slate-100/50" onclick="sort('t0', 2)">Exact Match Accuracy <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t0', 3)">Step-Level Accuracy <i class="fas fa-sort ml-1"></i></th>
</tr>
</thead>
<tbody class="divide-y divide-slate-100" id="t0-body"></tbody>
</table>
</div>
</div>
<div class="mt-4 text-slate-500 text-sm leading-relaxed bg-white/50 p-4 rounded-lg border border-slate-100">
<p><strong class="text-slate-700">Analysis:</strong> Models can often generate partially correct reasoning steps with moderate Step-Level Accuracy (SLA, up to ~60%) but fail to maintain end-to-end coherence, leading to low Exact-Match accuracy (~10-20% best cases); performance is especially weak on data and property questions where evidence is numerically dispersed across sources, with category accuracy rarely exceeding 30%, revealing failures in cross-paper numeric aggregation and mechanism synthesis rather than retrieval.</p>
</div>
</div>
<!-- Table 1: Idea Generation -->
<div class="mb-24 fade-in-up delay-100">
<div class="flex flex-col sm:flex-row justify-between items-end mb-4 gap-4">
<h3 class="text-2xl font-bold text-slate-800 flex items-center gap-2">
<span class="w-8 h-8 rounded-lg bg-purple-100 text-purple-600 flex items-center justify-center text-sm"><i class="fas fa-lightbulb"></i></span>
Idea Generation
</h3>
</div>
<div class="glass p-2 rounded-2xl hover:shadow-lg transition-shadow duration-300 mb-4 leaderboard-figure-wrap">
<img src="paper/imgs/idea_case.png" alt="Idea Generation Case" class="rounded-xl cursor-pointer hover:opacity-95 transition leaderboard-figure" onclick="openModal(this.src)" onerror="this.src='https://placehold.co/1200x500?text=Idea+Generation+Case'">
</div>
<div class="flex justify-center mb-3">
<div class="glass p-1 rounded-lg flex gap-2 text-sm w-fit overflow-x-auto">
<button onclick="filterTable('t1', 'all')" class="px-4 py-1.5 rounded-md bg-slate-900 text-white shadow transition-all filter-btn active whitespace-nowrap" data-t="t1" data-f="all">All Models</button>
<button onclick="filterTable('t1', 'open')" class="px-4 py-1.5 rounded-md text-slate-500 hover:text-slate-900 transition-all filter-btn whitespace-nowrap" data-t="t1" data-f="open">Open Source</button>
<button onclick="filterTable('t1', 'closed')" class="px-4 py-1.5 rounded-md text-slate-500 hover:text-slate-900 transition-all filter-btn whitespace-nowrap" data-t="t1" data-f="closed">Closed Source</button>
</div>
</div>
<div class="glass rounded-xl overflow-hidden ring-1 ring-slate-200 shadow-sm">
<div class="overflow-x-auto">
<table id="t1" class="w-full text-left text-sm whitespace-nowrap">
<thead>
<tr class="bg-slate-50/80 border-b border-slate-200 text-slate-500 font-semibold">
<th class="p-4 pl-6 cursor-pointer" onclick="sort('t1', 0)">Model <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t1', 1)">Type <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer bg-slate-100/50" onclick="sort('t1', 2)">Effectiveness <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t1', 3)">Novelty <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t1', 4)">Detailedness <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t1', 5)">Feasibility <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t1', 6)">Average <i class="fas fa-sort ml-1"></i></th>
</tr>
</thead>
<tbody class="divide-y divide-slate-100" id="t1-body"></tbody>
</table>
</div>
</div>
<div class="mt-4 text-slate-500 text-sm leading-relaxed bg-white/50 p-4 rounded-lg border border-slate-100">
<p><strong class="text-slate-700">Analysis:</strong> Models show high Novelty, especially closed-source like o3 and GPT-5, but Effectiveness is moderate and Feasibility remains low. Even top models with detailed ideas struggle to produce practical plans. Highest Feasibility scores are around 20–23, with open-source models generally lower despite decent Detailedness.</p>
</div>
</div>
<!-- Table 2: Dry Experiment Metrics -->
<div class="mb-24 fade-in-up delay-200">
<div class="flex flex-col sm:flex-row justify-between items-end mb-4 gap-4">
<h3 class="text-2xl font-bold text-slate-800 flex items-center gap-2">
<span class="w-8 h-8 rounded-lg bg-green-100 text-green-600 flex items-center justify-center text-sm"><i class="fas fa-code"></i></span>
Dry Experiment (Code Generation)
</h3>
</div>
<div class="glass p-2 rounded-2xl hover:shadow-lg transition-shadow duration-300 mb-4 leaderboard-figure-wrap">
<img src="paper/imgs/code_case1.png" alt="Dry Experiment Case" class="rounded-xl cursor-pointer hover:opacity-95 transition leaderboard-figure" onclick="openModal(this.src)" onerror="this.src='https://placehold.co/1200x500?text=Dry+Experiment+Case'">
</div>
<div class="flex justify-center mb-3">
<div class="glass p-1 rounded-lg flex gap-2 text-sm w-fit overflow-x-auto">
<button onclick="filterTable('t2', 'all')" class="px-4 py-1.5 rounded-md bg-slate-900 text-white shadow transition-all filter-btn active whitespace-nowrap" data-t="t2" data-f="all">All Models</button>
<button onclick="filterTable('t2', 'open')" class="px-4 py-1.5 rounded-md text-slate-500 hover:text-slate-900 transition-all filter-btn whitespace-nowrap" data-t="t2" data-f="open">Open Source</button>
<button onclick="filterTable('t2', 'closed')" class="px-4 py-1.5 rounded-md text-slate-500 hover:text-slate-900 transition-all filter-btn whitespace-nowrap" data-t="t2" data-f="closed">Closed Source</button>
</div>
</div>
<div class="glass rounded-xl overflow-hidden ring-1 ring-slate-200 shadow-sm">
<div class="overflow-x-auto">
<table id="t2" class="w-full text-left text-sm whitespace-nowrap">
<thead>
<tr class="bg-slate-50/80 border-b border-slate-200 text-slate-500 font-semibold">
<th class="p-4 pl-6 cursor-pointer" onclick="sort('t2', 0)">Model <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t2', 1)">Type <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer bg-slate-100/50" onclick="sort('t2', 2)">PassAll@5 <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t2', 3)">PassAll@3 <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t2', 4)">PassAll@1 <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t2', 5)">AET (s) <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t2', 6)">SER <i class="fas fa-sort ml-1"></i></th>
</tr>
</thead>
<tbody class="divide-y divide-slate-100" id="t2-body"></tbody>
</table>
</div>
</div>
<div class="mt-4 text-slate-500 text-sm leading-relaxed bg-white/50 p-4 rounded-lg border border-slate-100">
<p><strong class="text-slate-700">Analysis:</strong> Most models generate syntactically valid, runnable code (SER >90%, best 98.85) but show low test-case correctness when evaluated strictly (PassAll@1 ~42.07% → PassAll@5 ~36.64% best), and collapse on numerical integration and simulation fidelity, exposing that code fluency does not imply scientific computational competence, with numerical and physics-aware algorithm selection being the dominant failure mode.</p>
</div>
</div>
<!-- Table 3: Wet Experiment Metrics -->
<div class="mb-24 fade-in-up delay-300">
<div class="flex flex-col sm:flex-row justify-between items-end mb-4 gap-4">
<h3 class="text-2xl font-bold text-slate-800 flex items-center gap-2">
<span class="w-8 h-8 rounded-lg bg-teal-100 text-teal-600 flex items-center justify-center text-sm"><i class="fas fa-flask"></i></span>
Wet Experiment (Lab Protocol)
</h3>
</div>
<div class="glass p-2 rounded-2xl hover:shadow-lg transition-shadow duration-300 mb-4 leaderboard-figure-wrap">
<img src="paper/imgs/wet_case1.png" alt="Wet Experiment Case" class="rounded-xl cursor-pointer hover:opacity-95 transition leaderboard-figure" onclick="openModal(this.src)" onerror="this.src='https://placehold.co/1200x500?text=Wet+Experiment+Case'">
</div>
<div class="flex justify-center mb-3">
<div class="glass p-1 rounded-lg flex gap-2 text-sm w-fit overflow-x-auto">
<button onclick="filterTable('t3', 'all')" class="px-4 py-1.5 rounded-md bg-slate-900 text-white shadow transition-all filter-btn active whitespace-nowrap" data-t="t3" data-f="all">All Models</button>
<button onclick="filterTable('t3', 'open')" class="px-4 py-1.5 rounded-md text-slate-500 hover:text-slate-900 transition-all filter-btn whitespace-nowrap" data-t="t3" data-f="open">Open Source</button>
<button onclick="filterTable('t3', 'closed')" class="px-4 py-1.5 rounded-md text-slate-500 hover:text-slate-900 transition-all filter-btn whitespace-nowrap" data-t="t3" data-f="closed">Closed Source</button>
</div>
</div>
<div class="glass rounded-xl overflow-hidden ring-1 ring-slate-200 shadow-sm">
<div class="overflow-x-auto">
<table id="t3" class="w-full text-left text-sm whitespace-nowrap">
<thead>
<tr class="bg-slate-50/80 border-b border-slate-200 text-slate-500 font-semibold">
<th class="p-4 pl-6 cursor-pointer" onclick="sort('t3', 0)">Model <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t3', 1)">Type <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer bg-slate-100/50" onclick="sort('t3', 2)">Action Sequence Similarity <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t3', 3)">Parameter Accuracy <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t3', 4)">Average <i class="fas fa-sort ml-1"></i></th>
</tr>
</thead>
<tbody class="divide-y divide-slate-100" id="t3-body"></tbody>
</table>
</div>
</div>
<div class="mt-4 text-slate-500 text-sm leading-relaxed bg-white/50 p-4 rounded-lg border border-slate-100">
<p><strong class="text-slate-700">Analysis:</strong> LLMs can list reasonable lab actions but fail to organize them into correct experimental trajectories, showing uniformly low sequence similarity and parameter accuracy despite permutation equivalence checks, frequently adding irrelevant steps, omitting essential operations, or misordering branches, and struggling with temporal sampling, multi-sample pipeline alignment, and clear sample-purpose delineation.</p>
</div>
</div>
<!-- Table 4: Experimental Reasoning -->
<div class="mb-24 fade-in-up delay-300">
<div class="flex flex-col sm:flex-row justify-between items-end mb-4 gap-4">
<h3 class="text-2xl font-bold text-slate-800 flex items-center gap-2">
<span class="w-8 h-8 rounded-lg bg-pink-100 text-pink-600 flex items-center justify-center text-sm"><i class="fas fa-brain"></i></span>
Experimental Reasoning
</h3>
</div>
<div class="glass p-2 rounded-2xl hover:shadow-lg transition-shadow duration-300 mb-4 leaderboard-figure-wrap">
<img src="paper/imgs/mm_case_1.png" alt="Experimental Reasoning Case" class="rounded-xl cursor-pointer hover:opacity-95 transition leaderboard-figure" onclick="openModal(this.src)" onerror="this.src='https://placehold.co/1200x500?text=Experimental+Reasoning+Case'">
</div>
<div class="flex justify-center mb-3">
<div class="glass p-1 rounded-lg flex gap-2 text-sm w-fit overflow-x-auto">
<button onclick="filterTable('t4', 'all')" class="px-4 py-1.5 rounded-md bg-slate-900 text-white shadow transition-all filter-btn active whitespace-nowrap" data-t="t4" data-f="all">All Models</button>
<button onclick="filterTable('t4', 'open')" class="px-4 py-1.5 rounded-md text-slate-500 hover:text-slate-900 transition-all filter-btn whitespace-nowrap" data-t="t4" data-f="open">Open Source</button>
<button onclick="filterTable('t4', 'closed')" class="px-4 py-1.5 rounded-md text-slate-500 hover:text-slate-900 transition-all filter-btn whitespace-nowrap" data-t="t4" data-f="closed">Closed Source</button>
</div>
</div>
<div class="glass rounded-xl overflow-hidden ring-1 ring-slate-200 shadow-sm">
<div class="overflow-x-auto">
<table id="t4" class="w-full text-left text-sm whitespace-nowrap">
<thead>
<tr class="bg-slate-50/80 border-b border-slate-200 text-slate-500 font-semibold">
<th class="p-4 pl-6 cursor-pointer" onclick="sort('t4', 0)">Model <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t4', 1)">Type <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer bg-slate-100/50" onclick="sort('t4', 2)">Multi-choice Accuracy <i class="fas fa-sort ml-1"></i></th>
<th class="p-4 cursor-pointer" onclick="sort('t4', 3)">Reasoning Validity <i class="fas fa-sort ml-1"></i></th>
</tr>
</thead>
<tbody class="divide-y divide-slate-100" id="t4-body"></tbody>
</table>
</div>
</div>
<div class="mt-4 text-slate-500 text-sm leading-relaxed bg-white/50 p-4 rounded-lg border border-slate-100">
<p><strong class="text-slate-700">Analysis:</strong> Open-source models substantially lag behind closed ones in Multi-choice Accuracy (MCA), but most LLMs produce better Reasoning Validity than answer accuracy, achieving strong causal detection yet weak comparative reasoning, particularly on cross-image contrast, materials/life/Earth science domains, revealing persistent cognitive bottlenecks in subtle discrimination and numeric extraction over heterogeneous scientific visual evidence.</p>
</div>
</div>
</section>
<!-- ==================================================================================================== -->
<!-- 结论（Conclusion）板块：研究总结文本，仅展示内容，无需改逻辑。 -->
<!-- ==================================================================================================== -->
<!-- Conclusion -->
<section id="conclusion" class="py-2 px-4 max-w-4xl mx-auto">
<div class="glass rounded-2xl p-8 md:p-10 fade-in-up">
<h2 class="text-2xl font-bold text-slate-900 mb-6 text-center">Conclusion</h2>
<div class="text-slate-600 leading-relaxed space-y-4">
<p>
This work advances the study of Scientific General Intelligence (SGI) from both theory and practice. Grounded in the Practical Inquiry Model, we formalize SGI as the capacity to navigate the iterative cycle of Deliberation, Conception, Action, and Perception with the versatility of a human scientist. Building on this principle-grounded definition, we operationalize SGI through SGI-Bench, a comprehensive, scientist-aligned benchmark that instantiates four core task families: Scientific Deep Research, Idea Generation, Dry/Wet Experiment, and Experimental Reasoning. Complemented by our agentic evaluation framework and multi-metric protocol, SGI-Bench enables scalable, transparent, and domain-faithful assessment.
</p>
<p>
Experiments reveal a consistent pattern: in Deep Research, models show step-level alignment but low exact-match accuracy (10--20%), with brittleness in quantitative reasoning; in Idea Generation, hypotheses are fluent but underspecified and infeasible; in Dry Experiment, code is executable but PassAll@k remains low; in Wet Experiment, sequences show omissions and misordering; and in Experimental Reasoning, causal reasoning outperforms comparative, with persistent multimodal challenges. These highlight gaps between linguistic fluency and integrated scientific cognition. Moreover, SGI exhibits dynamic capacity: Test-Time Reinforcement Learning with novelty rewards improves idea generation without reference answers.
</p>
<p>
Taken together, SGI-Bench clarifies both what SGI is and where current systems fail. By integrating principled task design, multi-metric evaluation, and agentic tool use, our framework provides a concrete foundation for systematically advancing SGI. Looking forward, the combination of numerically robust reasoning, planning-aware conception, executable experimentation, comparative multimodal inference, dynamic test-time learning, and efficient tool ecosystems charts a clear path toward general intelligence systems capable of genuine scientific discovery.
</p>
</div>
</div>
</section>
<!-- ==================================================================================================== -->
<!-- 引用（Citation）板块：BibTeX 展示与复制按钮。复制功能由脚本实现。 -->
<!-- ==================================================================================================== -->
<!-- Citation -->
<section id="citation" class="py-8 px-4 max-w-4xl mx-auto">
<div class="glass rounded-2xl p-8 md:p-10 fade-in-up">
<h2 class="text-2xl font-bold text-slate-900 mb-6 text-center">Citation</h2>
<div class="relative bg-slate-900 rounded-lg group">
<div class="p-6 overflow-x-auto">
<pre class="text-slate-300 font-mono text-sm leading-relaxed"><code>@article{xu2025probing,
title={Probing Scientific General Intelligence of LLMs with Scientist-Aligned Workflows},
author={Xu, Wanghan and Zhou, Yuhao and Zhou, Yifan and Cao, Qinglong and Li, Shuo and Bu, Jia and Liu, Bo and Chen, Yixin and He, Xuming and Zhao, Xiangyu and others},
journal={arXiv preprint arXiv:2512.16969},
year={2025}
}</code></pre>
</div>
<button onclick="copyCitation()" class="absolute top-4 right-4 text-slate-500 hover:text-white transition-colors p-2"><i class="far fa-copy"></i></button>
</div>
</div>
</section>
<!-- ==================================================================================================== -->
<!-- 页脚（Footer）板块：版权信息与图标展示。 -->
<!-- ==================================================================================================== -->
<!-- Footer -->
<footer class="py-12 border-t border-slate-200 bg-white text-center">
<div class="flex justify-center gap-6 mb-4">
<a href="https://arxiv.org/pdf/2512.16969" target="_blank" rel="noopener noreferrer" class="hover:opacity-80 transition-opacity" title="arXiv">
<img src="images/arxiv.jpg" alt="arXiv" class="w-8 h-8 rounded-full border border-slate-200">
</a>
<a href="https://huggingface.co/collections/InternScience/sgi-bench" target="_blank" rel="noopener noreferrer" class="hover:opacity-80 transition-opacity" title="HuggingFace">
<img src="images/huggingface.jpg" alt="HuggingFace" class="w-8 h-8 rounded-full border border-slate-200">
</a>
<a href="https://github.com/InternScience/SGI-Bench" target="_blank" rel="noopener noreferrer" class="hover:opacity-80 transition-opacity" title="GitHub">
<img src="images/github.jpg" alt="GitHub" class="w-8 h-8 rounded-full border border-slate-200">
</a>
</div>
<p class="text-slate-500 text-sm">
© 2025 SGI-Bench.
Licensed under
<a href="https://github.com/InternScience/SGI-Bench/blob/main/LICENSE" target="_blank" rel="noopener noreferrer">
MIT
</a>.
</p>
<p class="text-slate-500 text-sm">
Contact:
<a href="https://black-yt.github.io/" target="_blank" rel="noopener noreferrer">
xu_wanghan@sjtu.edu.cn
</a>
</p>
<div class="flex justify-center mt-3">
<a href="https://raw.githubusercontent.com/InternScience/SGI-Bench/main/assets/wechat.jpg" target="_blank">
<img src="https://raw.githubusercontent.com/InternScience/SGI-Bench/main/assets/wechat.jpg"
alt="SGI-Bench WeChat QR Code"
class="w-[120px]">
</a>
</div>
</footer>
<!-- ==================================================================================================== -->
<!-- 灯箱（Lightbox Modal）板块：点击图片时全屏预览。淡入/淡出由 #lightbox 的过渡时间控制（见上方 CSS）。 -->
<!-- ==================================================================================================== -->
<!-- Lightbox Modal -->
<div id="lightbox" class="hidden fixed inset-0 z-[60] bg-black/90 backdrop-blur-sm flex items-center justify-center p-4" onclick="closeModal()">
<button class="absolute top-6 right-6 text-white/70 hover:text-white text-4xl focus:outline-none">×</button>
<!-- 灯箱图片最大高度由 max-h-[90vh] 控制（视口高度的 90%）；缩放过渡时长由 duration-300 控制（300ms） -->
<img id="lightbox-img" src="" class="max-w-full max-h-[90vh] rounded-lg shadow-2xl transform transition-transform duration-300 scale-95 object-contain" onclick="event.stopPropagation()">
</div>
<!-- ==================================================================================================== -->
<!-- 脚本（Scripts）板块：数据渲染、排序、UI 辅助、Demo 演示（包含可调常量）。 -->
<!-- 注意：如下所有注释仅为说明用途，未改变任何逻辑或文本内容。 -->
<!-- ==================================================================================================== -->
<script>
// =========================
// Data from sections/4-evaluation.tex (exact values)
// =========================
// Deep Research
const DEEP_DATA = [
// Open-source LLMs (前10个)
{ model: "DeepSeek-V3.2", type: "Open", em: 12.7, sla: 43.71 },
{ model: "DeepSeek-R1", type: "Open", em: 15.03, sla: 44.25 },
{ model: "Intern-S1", type: "Open", em: 15.74, sla: 56.83 },
{ model: "Intern-S1-mini", type: "Open", em: 11.06, sla: 49.30 },
{ model: "Kimi-k2", type: "Open", em: 13.11, sla: 52.64 },
{ model: "Qwen3-VL-235B-A22B", type: "Open", em: 11.97, sla: 53.85 },
{ model: "Qwen3-235B-A22B", type: "Open", em: 14.19, sla: 60.26 },
{ model: "Qwen3-8B", type: "Open", em: 8.18, sla: 44.3 },