-
Notifications
You must be signed in to change notification settings - Fork 8
Expand file tree
/
Copy pathmodel-plan.html
More file actions
526 lines (493 loc) · 45.2 KB
/
Copy pathmodel-plan.html
File metadata and controls
526 lines (493 loc) · 45.2 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
<!DOCTYPE html>
<html lang="zh-CN">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1, viewport-fit=cover">
<link rel="stylesheet" href="site-mobile.css">
<title>AI 模型旗舰横评对比 — 国内+海外 — AINav</title>
<meta name="description" content="2026-09 旗舰横评：Claude Opus 5.5 / Fable 5.1、GPT-6 Sol/Luna、Grok 4.7、Kimi K3、MiMo-V2.6、Hy4、GLM-5.3 / FlashX、DeepSeek-V4.1-Flash。决策模型见专题页。">
<link rel="canonical" href="https://aiv123.com/model-plan.html">
<meta property="og:type" content="website">
<meta property="og:title" content="AI 模型旗舰横评对比 — 国内+海外">
<meta property="og:description" content="2026-09 旗舰横评 · 含 Opus 5.5、GPT-6 Sol、Grok 4.7、MiMo-V2.6、DeepSeek Flash。">
<meta property="og:url" content="https://aiv123.com/model-plan.html">
<meta property="og:site_name" content="AINav">
<meta name="twitter:card" content="summary">
<link rel="icon" href="data:image/svg+xml,<svg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 100 100'><rect rx='18' width='100' height='100' fill='%230969da'/><text x='50' y='72' font-size='60' text-anchor='middle' fill='white' font-family='system-ui' font-weight='700'>AI</text></svg>">
<style>
:root, html[data-theme="dark"] {
--bg: #0f1419; --panel: #151b23; --card: #1c2430; --border: #2d3848;
--text: #e6edf3; --muted: #8b9cb3; --accent: #58a6ff; --accent2: #3fb950;
--warn: #d29922; --danger: #f85149; --hot: #ff6b6b; --overseas: #a78bfa;
}
html[data-theme="light"] {
--bg: #f6f8fa; --panel: #fff; --card: #fff; --border: #d0d7de;
--text: #1f2328; --muted: #59636e; --accent: #0969da; --accent2: #1a7f37;
--warn: #9a6700; --danger: #cf222e; --hot: #e53935; --overseas: #7c3aed;
}
@media (prefers-color-scheme: light) {
html[data-theme="system"] {
--bg: #f6f8fa; --panel: #fff; --card: #fff; --border: #d0d7de;
--text: #1f2328; --muted: #59636e; --accent: #0969da; --accent2: #1a7f37;
--warn: #9a6700; --danger: #cf222e; --hot: #e53935; --overseas: #7c3aed;
}
}
* { box-sizing: border-box; }
body { margin: 0; font-family: "Segoe UI", system-ui, "PingFang SC", "Microsoft YaHei", sans-serif;
background: var(--bg); color: var(--text); line-height: 1.6; }
a { color: var(--accent); text-decoration: none; }
a:hover { text-decoration: underline; }
.cp-top { padding: 0.75rem 1rem; border-bottom: 1px solid var(--border); background: var(--panel); position: sticky; top: 0; z-index: 10; }
.cp-top-inner { max-width: 76rem; margin: 0 auto; display: flex; flex-wrap: wrap; align-items: center; gap: 0.5rem 1rem; justify-content: space-between; }
.cp-nav a { font-weight: 600; margin-right: 1rem; font-size: 0.88rem; }
.cp-toolbar { display: flex; flex-wrap: wrap; gap: 0.35rem; align-items: center; }
.theme-btn { font-size: 0.72rem; padding: 0.2rem 0.5rem; border-radius: 5px; border: 1px solid var(--border);
background: var(--card); color: var(--muted); cursor: pointer; font-family: inherit; }
.theme-btn.is-active { color: var(--accent); border-color: var(--accent); font-weight: 600; }
.cp-wrap { max-width: 76rem; margin: 0 auto; padding: 1rem; }
h1 { font-size: 1.35rem; margin: 0 0 0.3rem; }
.cp-subtitle { font-size: 0.88rem; color: var(--muted); margin-bottom: 1rem; }
.cp-section-hd { font-size: 1.1rem; font-weight: 700; margin: 1.8rem 0 0.8rem; padding-bottom: 0.3rem; border-bottom: 2px solid var(--border); display: flex; align-items: center; gap: 0.5rem; }
.cp-section-hd.overseas-hd { border-bottom-color: var(--overseas); }
.cp-legend { display: flex; flex-wrap: wrap; gap: 0.8rem; margin-bottom: 1.2rem; font-size: 0.82rem; }
.cp-badge { display: inline-block; padding: 0.1rem 0.45rem; border-radius: 4px; font-size: 0.72rem; font-weight: 600; }
.badge-deepseek { background: rgba(88,166,255,0.15); color: var(--accent); border: 1px solid rgba(88,166,255,0.3); }
.badge-glm { background: rgba(63,185,80,0.15); color: var(--accent2); border: 1px solid rgba(63,185,80,0.3); }
.badge-kimi { background: rgba(249,115,22,0.15); color: #f97316; border: 1px solid rgba(249,115,22,0.3); }
.badge-qwen { background: rgba(168,85,247,0.15); color: #a855f7; border: 1px solid rgba(168,85,247,0.3); }
.badge-minimax { background: rgba(236,72,153,0.15); color: #ec4899; border: 1px solid rgba(236,72,153,0.3); }
.badge-gemma { background: rgba(34,197,94,0.15); color: #22c55e; border: 1px solid rgba(34,197,94,0.3); }
.badge-gpt { background: rgba(88,166,255,0.15); color: var(--accent); border: 1px solid rgba(88,166,255,0.3); }
.badge-claude { background: rgba(210,153,34,0.15); color: var(--warn); border: 1px solid rgba(210,153,34,0.3); }
.badge-gemini { background: rgba(63,185,80,0.15); color: var(--accent2); border: 1px solid rgba(63,185,80,0.3); }
.badge-overseas { background: rgba(167,139,250,0.15); color: var(--overseas); border: 1px solid rgba(167,139,250,0.3); }
.badge-hot { background: rgba(255,107,107,0.15); color: var(--hot); border: 1px solid rgba(255,107,107,0.3); }
.badge-hy { background: rgba(0,82,217,0.15); color: #3b82f6; border: 1px solid rgba(0,82,217,0.3); }
.cp-platforms { display: grid; grid-template-columns: 1fr; gap: 1.2rem; }
@media (min-width: 52rem) { .cp-platforms { grid-template-columns: 1fr 1fr; } }
.cp-platform { border: 1px solid var(--border); border-radius: 10px; background: var(--card); overflow: hidden; }
.cp-platform.overseas { border-color: rgba(167,139,250,0.3); }
.cp-platform-head { padding: 0.75rem 1rem; border-bottom: 1px solid var(--border); display: flex; flex-wrap: wrap; align-items: center; gap: 0.5rem; }
.cp-platform-name { font-size: 1.05rem; font-weight: 700; }
.cp-platform-name a { color: var(--text); }
.cp-platform-body { padding: 0.75rem 1rem; font-size: 0.85rem; line-height: 1.7; }
.tech-tag { display: inline-block; padding: 0.05rem 0.4rem; border-radius: 4px; font-size: 0.72rem; font-weight: 600; margin: 0.15rem; background: rgba(88,166,255,0.08); border: 1px solid rgba(88,166,255,0.15); color: var(--accent); }
.good { color: var(--accent2); font-weight: 600; }
.bad { color: var(--danger); opacity: 0.85; }
.cp-scenario { margin-top: 1rem; border: 1px solid var(--border); border-radius: 10px; background: var(--card); overflow: hidden; }
.cp-scenario-head { padding: 0.65rem 1rem; border-bottom: 1px solid var(--border); font-weight: 700; font-size: 0.95rem; }
.cp-scenario-body { padding: 0.75rem 1rem; font-size: 0.85rem; line-height: 1.75; }
.candidate { color: var(--muted); font-size: 0.82rem; }
.recommend { color: var(--accent2); font-weight: 700; font-size: 0.95rem; }
.reason { margin-top: 0.4rem; padding-left: 1rem; border-left: 3px solid var(--accent2); }
.cp-matrix-wrap { overflow-x: auto; border: 1px solid var(--border); border-radius: 8px; background: var(--card); margin-top: 1rem; }
.cp-matrix { width: 100%; border-collapse: collapse; font-size: 0.8rem; }
.cp-matrix th, .cp-matrix td { padding: 0.45rem 0.65rem; border: 1px solid var(--border); text-align: left; }
.cp-matrix th { background: var(--panel); font-weight: 600; color: var(--muted); font-size: 0.76rem; }
.cp-matrix .model-name { font-weight: 700; white-space: nowrap; }
.cp-summary { margin-top: 2rem; padding: 1rem 1.2rem; border: 2px solid var(--accent); border-radius: 10px; background: rgba(88,166,255,0.04); }
.cp-summary h2 { font-size: 1.05rem; margin: 0 0 0.8rem; }
.cp-summary-grid { display: grid; grid-template-columns: 1fr; gap: 0.5rem; font-size: 0.88rem; }
@media (min-width: 40rem) { .cp-summary-grid { grid-template-columns: 1fr 1fr; } }
.cp-summary-item { display: flex; align-items: center; gap: 0.5rem; padding: 0.4rem 0.6rem; border-radius: 6px; background: var(--card); border: 1px solid var(--border); }
.cp-summary-item .arrow { color: var(--accent2); font-weight: 700; }
.cp-updated-line { font-size: 0.76rem; color: var(--muted); margin: 0 0 0.3rem; }
.cp-footer { max-width: 76rem; margin: 2rem auto 1rem; padding: 0 1rem; font-size: 0.76rem; color: var(--muted); text-align: center; }
</style>
<script async src="https://www.googletagmanager.com/gtag/js?id=G-2B8FBWRX4N"></script>
<script>window.dataLayer=window.dataLayer||[];function gtag(){dataLayer.push(arguments);}gtag('js',new Date());gtag('config','G-2B8FBWRX4N');</script>
<script defer src="https://static.cloudflareinsights.com/beacon.min.js" data-cf-beacon='{"token": "2d49f38b39c743ac80b6d6b9ff99494d"}'></script>
</head>
<body>
<div class="cp-top">
<div class="cp-top-inner">
<div class="cp-nav" data-plan-nav></div>
<div class="cp-toolbar">
<button type="button" class="theme-btn" data-theme-set="system">跟随系统</button>
<button type="button" class="theme-btn" data-theme-set="light">浅色</button>
<button type="button" class="theme-btn" data-theme-set="dark">深色</button>
</div>
</div>
</div>
<div class="cp-wrap">
<div style="margin-bottom:1rem;padding:.55rem .85rem;border-radius:8px;border:1px solid var(--border);background:var(--card);display:flex;flex-wrap:wrap;align-items:center;justify-content:center;gap:.35rem .6rem;font-size:.84rem;line-height:1.5;">
<strong style="color:var(--accent2)">合作推广</strong> <span style="color:var(--muted)">·</span> <span style="color:var(--text)">GPT / Claude / Gemini</span> <span style="color:var(--muted)">·</span> <span style="color:var(--text)">100+ 模型</span> <span style="color:var(--muted)">·</span> <span style="color:var(--text)">文本 / 图像 / 视频</span> <span style="color:var(--muted)">·</span> <span style="color:var(--text)">支持缓存</span> <span style="color:var(--muted)">·</span> <span style="color:var(--text)">不留存对话</span> <span style="color:var(--muted)">·</span> <span style="color:var(--accent2);font-weight:600">首充赠 $3</span> <a href="https://ofox.io/x/aiv123" target="_blank" rel="noopener noreferrer sponsored" style="display:inline-flex;align-items:center;font-size:.78rem;font-weight:600;color:var(--accent2);padding:.2rem .55rem;border-radius:5px;border:1px solid rgba(63,185,80,.4);background:rgba(63,185,80,.1);text-decoration:none;white-space:nowrap;">立即注册</a>
</div>
<h1>🧠 AI 模型选型横评</h1>
<p class="cp-subtitle">2026-09 旗舰：Claude Opus 5.5 / Fable 5.1、GPT-6 Sol / Luna、Grok 4.7、Kimi K3、MiMo-V2.6、Step 5 Preview、Hy4、GLM-5.3 / FlashX、<strong>DeepSeek-V4.1-Flash</strong>（API <code>deepseek-flash</code>）。要分类/路由/打分这类<strong>非聊天</strong>决策，另见 <a href="decision-model-plan.html">决策模型</a>。Agent 名次对照 <a href="https://arena.ai/leaderboard/agent" target="_blank" rel="noopener noreferrer">Arena Agent 榜</a>（2026-09-05）。</p>
<p class="cp-updated-line"><span class="cp-updated">最后更新：2026-09-23</span> · 刊例以各厂商价表为准</p>
<p style="margin:0 0 1rem;font-size:.88rem;">不确定选哪个？先用 <a href="model-finder.html" style="font-weight:700;">模型推荐器</a> 答 3 个问题，再回来看横评。本机跑权重（Ollama / 显存 / GGUF）看 <a href="local-model-plan.html" style="font-weight:700;">本地模型</a>。</p>
<div class="cp-legend">
<span class="cp-legend-item"><span class="cp-badge badge-deepseek">DeepSeek</span> 长上下文</span>
<span class="cp-legend-item"><span class="cp-badge badge-glm">GLM-5.3</span> 长程 Coding</span>
<span class="cp-legend-item"><span class="cp-badge badge-kimi">Kimi</span> Agent</span>
<span class="cp-legend-item"><span class="cp-badge badge-hy">Hy4</span> 开源 Agent</span>
<span class="cp-legend-item"><span class="cp-badge badge-qwen">Qwen 3.8</span> 开源旗舰</span>
<span class="cp-legend-item"><span class="cp-badge badge-minimax">MiniMax</span> 速度</span>
<span class="cp-legend-item"><span class="cp-badge badge-gemma">Gemma</span> 端侧</span>
<span class="cp-legend-item"><span class="cp-badge badge-overseas">Fable 5.1 / Opus 5.5 / GPT-6 / Grok 4.7</span> 海外旗舰</span>
<span class="cp-legend-item"><a href="decision-model-plan.html" class="cp-badge badge-new" style="text-decoration:none;">决策模型</a> Jev / Laya</span>
</div>
<div class="cp-section-hd"><span class="cp-flag">🇨🇳</span> 国内旗舰模型</div>
<div class="cp-platforms">
<div class="cp-platform">
<div class="cp-platform-head">
<span class="cp-platform-name"><a href="https://api-docs.deepseek.com/news/news260910" target="_blank" rel="noopener noreferrer">DeepSeek-V4.1-Flash</a></span>
<span class="cp-badge badge-deepseek">速度 / 多模态</span>
<span class="cp-badge badge-hot">🆕 9/10</span>
</div>
<div class="cp-platform-body">
<strong>核心技术：</strong><span class="tech-tag">552B MoE</span><span class="tech-tag">非对称 Encoder-Decoder</span><span class="tech-tag">原生视觉</span><span class="tech-tag">API <code>deepseek-flash</code></span><br>
<strong class="good">✓ 天然适合：</strong>便宜快调用、编程 / Agent 轻量循环、图文理解、Strix 一类<strong>快扫</strong><br>
<strong class="bad">✗ 天然不适合：</strong>当成「深扫最强」；端侧手机（这是云 API / 大 MoE）<br>
<strong>设计理念：</strong>2026-09-10 上线，官方称输入激活约 8B、输出约 16B。新接入一律用 <code>deepseek-flash</code>。旧 ID <code>deepseek-v4-flash</code> 等兼容转发到 V4.1-Flash。官方说明：自 <strong>2026-09-14 04:00 UTC</strong> 起，<code>deepseek-v4-pro</code> 请求暂时也转发到 V4.1-Flash（按 Flash 计费），直到 V4.1-Pro 上线——选型别把「Pro」字样当仍可用的独立强档
</div>
</div>
<div class="cp-platform">
<div class="cp-platform-head">
<span class="cp-platform-name"><a href="https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" target="_blank" rel="noopener noreferrer">DeepSeek-V4-Pro</a></span>
<span class="cp-badge badge-deepseek">长上下文 · 过渡</span>
</div>
<div class="cp-platform-body">
<strong>核心技术：</strong><span class="tech-tag">1.6T MoE</span><span class="tech-tag">CSA+HCA</span><span class="tech-tag">1M 上下文</span><span class="tech-tag">Think Max</span><br>
<strong class="good">✓ 天然适合：</strong>开源权重研究复现、历史文档仍写 Pro 的长上下文场景（以你实际拿到的推理端为准）<br>
<strong class="bad">✗ 天然不适合：</strong>默认假设官网 API 的 <code>deepseek-v4-pro</code> 仍是独立 Pro 权重——见上条 Flash 转发说明<br>
<strong>设计理念：</strong>稀疏注意力把 1M 做成「可用」。权重仍在 HF；云 API 侧以官方 changelog 为准。Strix 深扫更稳的默认仍是 <code>glm-5.3</code>，别死磕已转发的 Pro ID
</div>
</div>
<div class="cp-platform">
<div class="cp-platform-head">
<span class="cp-platform-name"><a href="https://z.ai/blog/glm-5.3" target="_blank" rel="noopener noreferrer">GLM-5.3</a></span>
<span class="cp-badge badge-glm">长程 Coding</span>
<span class="cp-badge badge-hot">🆕 8/14</span>
</div>
<div class="cp-platform-body">
<strong>核心技术：</strong><span class="tech-tag">同 5.2 底座</span><span class="tech-tag">后训练加码</span><span class="tech-tag">常开思考</span><span class="tech-tag">Coding Plan 全量</span><br>
<strong class="good">✓ 天然适合：</strong>项目级工程、终端 Agent、Strix 默认深扫、对标 Opus 档的国产长程 Coding<br>
<strong class="bad">✗ 天然不适合：</strong>关掉思考的极低延迟闲聊（思考不可关，只能调 low/high/max）<br>
<strong>设计理念：</strong>底座仍是 GLM-5.2；8/14 只加后训练。Coding Plan 用户已切到 5.3。速度档另见下方 FlashX
</div>
</div>
<div class="cp-platform">
<div class="cp-platform-head">
<span class="cp-platform-name"><a href="https://z.ai/" target="_blank" rel="noopener noreferrer">GLM-5.3-Flash / FlashX</a></span>
<span class="cp-badge badge-glm">速度档</span>
<span class="cp-badge badge-hot">🆕 9/18</span>
</div>
<div class="cp-platform-body">
<strong>核心技术：</strong><span class="tech-tag">原生多模态 Flash</span><span class="tech-tag">FlashX 高吞吐</span><span class="tech-tag">API 已上线</span><br>
<strong class="good">✓ 天然适合：</strong>要快、要控成本的浅扫 / 高并发；FlashX 官方称最高约 <strong>200 tokens/s</strong>（ainews / AIBase 2026-09-18）<br>
<strong class="bad">✗ 天然不适合：</strong>顶替 GLM-5.3 全量深扫与长程工程；价格刊例以智谱官方为准，本站不编<br>
<strong>设计理念：</strong>Flash 曾以 Ox Alpha 匿名亮相；FlashX 是同线提速版。Strix 更省档可写 <code>glm-5.3-flash</code>，默认深扫仍用 <code>glm-5.3</code>
</div>
</div>
<div class="cp-platform">
<div class="cp-platform-head">
<span class="cp-platform-name"><a href="https://www.kimi.com/zh-cn/blog/kimi-k3" target="_blank" rel="noopener noreferrer">Kimi K3</a></span>
<span class="cp-badge badge-kimi">Agent 深度</span>
<span class="cp-badge badge-new">🆕 2.8T</span>
</div>
<div class="cp-platform-body">
<strong>核心技术：</strong><span class="tech-tag">2.8T 开源</span><span class="tech-tag">1M 上下文</span><span class="tech-tag">原生多模态</span><span class="tech-tag">Arena Agent #7</span><br>
<strong class="good">✓ 天然适合：</strong>长仓工程、多步 Agent、知识工作与深度推理<br>
<strong class="bad">✗ 天然不适合：</strong>极致低价批量（可用 K2.x / Flash 档）<br>
<strong>设计理念：</strong>开源旗舰；Kimi Code / Work / API（<code>kimi-k3</code>）可用。2026-09 起有 <strong>Kimi Code 桌面版</strong>（macOS / Windows）。Arena Agent（2026-09-05）K3 Max 第 7
</div>
</div>
<div class="cp-platform">
<div class="cp-platform-head">
<span class="cp-platform-name"><a href="https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL" target="_blank" rel="noopener noreferrer">Xiaomi MiMo-V2.6</a></span>
<span class="cp-badge badge-hot">🆕 9/22</span>
<span class="cp-badge badge-qwen">开源全模态</span>
</div>
<div class="cp-platform-body">
<strong>核心技术：</strong><span class="tech-tag">Pro / Flash</span><span class="tech-tag">AA 指数约 46</span><span class="tech-tag">HF 开源</span><br>
<strong class="good">✓ 天然适合：</strong>要开源全模态底座、Agent 基准接近 Opus 5 / 上一代 Sol 档<br>
<strong class="bad">✗ 天然不适合：</strong>对标 Fable 5.1 / GPT-6 Astra；Token Plan 刊例仍以官网 V2.5 为准<br>
<strong>设计理念：</strong>2026-09-22 开源 Pro/Flash。多家评测称 AA 综合智能指数约 46，开源侧很靠前，仍低于顶级闭源。权重见 Hugging Face XiaomiMiMo
</div>
</div>
<div class="cp-platform">
<div class="cp-platform-head">
<span class="cp-platform-name"><a href="https://platform.stepfun.com/" target="_blank" rel="noopener noreferrer">Step 5 Preview</a></span>
<span class="cp-badge badge-new">🆕 9/20</span>
</div>
<div class="cp-platform-body">
<strong>核心技术：</strong><span class="tech-tag">600B MoE</span><span class="tech-tag">激活 ~27B</span><span class="tech-tag">1M 上下文</span><span class="tech-tag">原生多模态</span><br>
<strong class="good">✓ 天然适合：</strong>长程 Agentic 任务、要开源性价比逼近 K3 档<br>
<strong class="bad">✗ 天然不适合：</strong>把 Preview 当最终 GA；完整权重节奏以阶跃公告为准<br>
<strong>设计理念：</strong>阶跃星辰 2026-09-20 起 API / Studio 可用。套餐池见 <a href="coding-plan.html">编程套餐</a> Step Plan
</div>
</div>
<div class="cp-platform">
<div class="cp-platform-head">
<span class="cp-platform-name"><a href="https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421" target="_blank" rel="noopener noreferrer">Qwen 3.8 Max</a></span>
<span class="cp-badge badge-qwen">开源旗舰</span>
<span class="cp-badge badge-hot">🆕 8/3</span>
</div>
<div class="cp-platform-body">
<strong>核心技术：</strong><span class="tech-tag">2.4T MoE</span><span class="tech-tag">Max 级开源</span><span class="tech-tag">1M 上下文</span><span class="tech-tag">$2/$6</span><br>
<strong class="good">✓ 天然适合：</strong>编码/Cowork、长程任务、要开源权重的旗舰替代<br>
<strong class="bad">✗ 天然不适合：</strong>一张 24GB 卡（那是 <a href="local-model-plan.html">Qwen3.8-27B</a>）；纯图便宜批量可仍看 3.6 API<br>
<strong>设计理念：</strong>第一次把 Max 档权重开源；API 约 $2/$6 每百万 token。消费级本地请走 27B，不要找 Max 的 GGUF。Arena Agent #16
</div>
</div>
<div class="cp-platform">
<div class="cp-platform-head">
<span class="cp-platform-name"><a href="https://huggingface.co/tencent/Hy4-preview" target="_blank" rel="noopener noreferrer">Tencent Hy4 preview</a></span>
<span class="cp-badge badge-hy">开源 Agent</span>
<span class="cp-badge badge-hot">🆕 8/28</span>
</div>
<div class="cp-platform-body">
<strong>核心技术：</strong><span class="tech-tag">770B MoE</span><span class="tech-tag">49B 激活</span><span class="tech-tag">1M 上下文</span><span class="tech-tag">Apache 2.0</span><span class="tech-tag">$0.83/$2.50</span><br>
<strong class="good">✓ 天然适合：</strong>国内直连的开源 Agent、编码/办公、要低价长上下文的生产力任务<br>
<strong class="bad">✗ 天然不适合：</strong>把 preview 当最终 GA；端侧单卡（这是云端 MoE）<br>
<strong>设计理念：</strong>混元 Hy 线预览权重。Arena Agent（2026-09-05）排第 10，任务成本中位约 $0.27。API / 元宝 / CodeBuddy / OpenRouter 可调用；权重在 Hugging Face
</div>
</div>
<div class="cp-platform">
<div class="cp-platform-head">
<span class="cp-platform-name"><a href="https://www.minimaxi.com" target="_blank" rel="noopener noreferrer">MiniMax M2.7</a></span>
<span class="cp-badge badge-minimax">速度</span>
</div>
<div class="cp-platform-body">
<strong>核心技术：</strong><span class="tech-tag">10B 激活</span><span class="tech-tag">30 tok/s</span><span class="tech-tag">流式低延迟</span><br>
<strong class="good">✓ 天然适合：</strong>实时对话、客服/语音、高并发 API、流式交互<br>
<strong class="bad">✗ 天然不适合：</strong>复杂长文档深度推理、顶级代码架构设计<br>
<strong>设计理念：</strong>速度是第一要素，交互体验优先于极限智商
</div>
</div>
<div class="cp-platform">
<div class="cp-platform-head">
<span class="cp-platform-name"><a href="https://ai.google.dev/gemma" target="_blank" rel="noopener noreferrer">Gemma 4</a></span>
<span class="cp-badge badge-gemma">端侧</span>
</div>
<div class="cp-platform-body">
<strong>核心技术：</strong><span class="tech-tag">E2B 端侧</span><span class="tech-tag">PLE 多模态</span><span class="tech-tag">140+ 语言</span><span class="tech-tag">Apache 2.0</span><br>
<strong class="good">✓ 天然适合：</strong>端侧 AI 助手、隐私本地、离线翻译、IoT<br>
<strong class="bad">✗ 天然不适合：</strong>超大模型搜索、1M 上下文长任务<br>
<strong>设计理念：</strong>端侧 AI 的「平台级能力」，车机/IoT/手机端为主战场。手机运行时与能干活小模型见 <a href="edge-model-plan.html">端侧模型</a>；桌面 GGUF / 显存分档见 <a href="local-model-plan.html">本地模型横评</a>
</div>
</div>
</div>
<div class="cp-section-hd overseas-hd"><span class="cp-flag">🌍</span> 海外旗舰模型</div>
<div class="cp-platforms">
<div class="cp-platform overseas">
<div class="cp-platform-head">
<span class="cp-platform-name"><a href="https://openai.com/index/introducing-gpt-6-sol-and-luna" target="_blank" rel="noopener noreferrer">GPT-6 Sol / Luna (OpenAI)</a></span>
<span class="cp-badge badge-gpt">通用旗舰</span>
<span class="cp-badge badge-hot">🆕 9/23</span>
<span class="cp-badge badge-overseas">海外</span>
</div>
<div class="cp-platform-body">
<strong>核心技术：</strong><span class="tech-tag">Sol / Luna</span><span class="tech-tag">对齐 Astra 训练法</span><span class="tech-tag">日常工作降本</span><br>
<strong class="good">✓ 天然适合：</strong>ChatGPT 日常付费默认、要接近旗舰但更省钱的编程/办公/Agent<br>
<strong class="bad">✗ 天然不适合：</strong>无妥协极限仍选 <strong>GPT-6 Astra</strong>；国内直连需网关<br>
<strong>设计理念：</strong>2026-09-23 发布。OpenAI 称 Astra 仍是整体最强；Sol / Luna 把接近旗舰的能力放到更快、更省的尺寸。报道称 API 价较 GPT-5.6 促销价约降一半——<strong>绝对刊例以 OpenAI 价表为准</strong>，本站不写死
</div>
</div>
<div class="cp-platform overseas">
<div class="cp-platform-head">
<span class="cp-platform-name"><a href="https://www.anthropic.com/claude-fable-and-mythos-5-1" target="_blank" rel="noopener noreferrer">Claude Fable 5.1</a></span>
<span class="cp-badge badge-claude">长程旗舰</span>
<span class="cp-badge badge-overseas">海外</span>
</div>
<div class="cp-platform-body">
<strong>核心技术：</strong><span class="tech-tag">1M 上下文</span><span class="tech-tag">128K 输出</span><span class="tech-tag">缓存读 $0.25</span><span class="tech-tag">Arena Agent #1</span><br>
<strong class="good">✓ 天然适合：</strong>数小时/数天异步长程任务、Agent 编排、复杂知识工作；<a href="https://arena.ai/leaderboard/agent" target="_blank" rel="noopener noreferrer">Arena Agent</a> 总分第一（2026-09-05）<br>
<strong class="bad">✗ 天然不适合：</strong>日常编码（先用 Opus 5.5）；生科/攻防网安仍受安全分类器约束，可信访问走 Mythos 5.1<br>
<strong>设计理念：</strong>9/1 接档 Fable 5。刊例仍 <strong>$10/$50</strong>，缓存命中 <strong>$0.25</strong>。日常稳妥档优先 Opus 5.5
</div>
</div>
<div class="cp-platform overseas">
<div class="cp-platform-head">
<span class="cp-platform-name"><a href="https://www.anthropic.com/news/claude-opus-5" target="_blank" rel="noopener noreferrer">Claude Opus 5.5</a></span>
<span class="cp-badge badge-claude">深度推理</span>
<span class="cp-badge badge-hot">🆕 9/23</span>
<span class="cp-badge badge-overseas">海外</span>
</div>
<div class="cp-platform-body">
<strong>核心技术：</strong><span class="tech-tag">Claude 5.5 首款</span><span class="tech-tag">近 Fable 5.1</span><span class="tech-tag">成本↓约40%</span><span class="tech-tag">速度↑约30%</span><br>
<strong class="good">✓ 天然适合：</strong>日常长程 Coding / Claude Code 主力；要接近 Fable 智商但更省<br>
<strong class="bad">✗ 天然不适合：</strong>中文略弱、需梯子、端侧；绝对刊例以 Anthropic 价表为准<br>
<strong>设计理念：</strong>2026-09-23 上线 Claude 与主要云。官方称多数任务接近 Fable 5.1，典型任务成本较 Opus 5 约降 40%、输出约提速 30%。刊例以 Anthropic 为准；上代 Opus 5（$5/$25）可作对照
</div>
</div>
<div class="cp-platform overseas">
<div class="cp-platform-head">
<span class="cp-platform-name"><a href="https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/" target="_blank" rel="noopener noreferrer">Gemini 3.8 Flash</a></span>
<span class="cp-badge badge-gemini">速度/Agent</span>
<span class="cp-badge badge-overseas">海外</span>
</div>
<div class="cp-platform-body">
<strong>核心技术：</strong><span class="tech-tag">$0.75/$3.75 引入至 12/31</span><span class="tech-tag">长程工程</span><span class="tech-tag">2027-01-01 翻倍</span><br>
<strong class="good">✓ 天然适合：</strong>软件工程、多步 Agent、要便宜速度的 Google 生态<br>
<strong class="bad">✗ 天然不适合：</strong>把 Flash 当 Pro 用；引入价只到 <strong>2026-12-31</strong>，次日变 $1.50/$7.50<br>
<strong>设计理念：</strong>接档 3.7，同价同速、推理与编码加码。ID <code>gemini-3.8-flash</code>。1M 研究/视频理解仍可看 3.1 Pro
</div>
</div>
<div class="cp-platform overseas">
<div class="cp-platform-head">
<span class="cp-platform-name"><a href="https://x.ai/" target="_blank" rel="noopener noreferrer">Grok 4.7 (SpaceXAI / xAI)</a></span>
<span class="cp-badge badge-overseas">海外</span>
<span class="cp-badge badge-hot">🆕 9/21</span>
</div>
<div class="cp-platform-body">
<strong>核心技术：</strong><span class="tech-tag">~2.1T</span><span class="tech-tag">$2/$6</span><span class="tech-tag">长任务</span><span class="tech-tag">编程/知识工作</span><br>
<strong class="good">✓ 天然适合：</strong>写代码、数小时级长跑 Agent、要高性价比旗舰<br>
<strong class="bad">✗ 天然不适合：</strong>对标 Claude Fable 5.1 / GPT-6 Astra 的极限档；长上下文溢价规则以 xAI 价表为准<br>
<strong>设计理念：</strong>2026-09-21/22 上线。主打编程与知识工作长任务；报道称速度约同类两倍、价格约旗舰一半，API 约 $2/$6 每百万 token（与 4.6 刊例同量级）。综合仍逊于 Fable 5.1 / Astra
</div>
</div>
</div>
<div class="cp-section-hd" id="decision"><span class="cp-flag">⚡</span> 决策模型（非聊天）</div>
<p style="font-size:.85rem;line-height:1.7;margin:0 0 1rem;padding:.75rem 1rem;border:1px solid var(--border);border-radius:10px;background:var(--card);">
分类、路由、打分、是否分支这类<strong>给程序用的结构化决策</strong>，不要和聊天旗舰混选。闭源云 API 看 TypeSafe <strong>Jev</strong>；开源权重看 <strong>Laya</strong>（Apache 2.0）等。详见专题 <a href="decision-model-plan.html" style="font-weight:700;">决策模型</a>。
</p>
<div class="cp-section-hd"><span class="cp-flag">📊</span> 技术特性 × 场景速查表</div>
<div class="cp-matrix-wrap">
<table class="cp-matrix">
<thead>
<tr><th>模型</th><th>核心技术亮点</th><th>天然适合</th><th>天然不适合</th></tr>
</thead>
<tbody>
<tr><td class="model-name">DeepSeek-V4.1-Flash</td><td>552B MoE · API <code>deepseek-flash</code> · 原生视觉</td><td class="good">便宜快、轻量 Agent、快扫</td><td class="bad">深扫最强、端侧</td></tr>
<tr><td class="model-name">DeepSeek-V4-Pro</td><td>权重仍在 · API ID 过渡期可能转发 Flash</td><td class="good">开源复现、历史长上下文文档</td><td class="bad">默认当独立云端强档</td></tr>
<tr><td class="model-name">GLM-5.3</td><td>5.2 底座 · 后训练 · 常开思考 · Coding Plan</td><td class="good">项目级工程、终端 Agent、Strix 默认</td><td class="bad">思考关不掉</td></tr>
<tr><td class="model-name">GLM-5.3-Flash / FlashX</td><td>速度档 · FlashX ~200 tok/s（报道）</td><td class="good">浅扫、高吞吐</td><td class="bad">顶替 5.3 深扫</td></tr>
<tr><td class="model-name">Kimi K3</td><td>2.8T · 1M · Arena Agent #7 · Code Desktop</td><td class="good">复杂工程、多步工作流</td><td class="bad">纯单轮低价批量</td></tr>
<tr><td class="model-name">MiMo-V2.6</td><td>开源全模态 · AA 46 · Pro/Flash</td><td class="good">最强开源全模态底座</td><td class="bad">闭源极限档</td></tr>
<tr><td class="model-name">Step 5 Preview</td><td>600B MoE · ~27B 激活 · 1M</td><td class="good">Agentic 长任务性价比</td><td class="bad">Preview / 权重节奏</td></tr>
<tr><td class="model-name">Qwen 3.8 Max</td><td>2.4T · Max 开源 · $2/$6</td><td class="good">编码/Cowork、开源旗舰</td><td class="bad">单卡请改 27B</td></tr>
<tr><td class="model-name">Hy4 preview</td><td>770B MoE · Apache 2.0 · $0.83/$2.50</td><td class="good">国内开源 Agent、低价长上下文</td><td class="bad">preview，非端侧</td></tr>
<tr><td class="model-name">MiniMax M2.7</td><td>10B 激活 · 30 tok/s · 流式</td><td class="good">实时对话、高并发 API</td><td class="bad">超长文档深度推理</td></tr>
<tr><td class="model-name">Gemma 4</td><td>E2B 端侧 · PLE · 140+ 语言 · Apache 2.0</td><td class="good">端侧/离线/隐私</td><td class="bad">1M 长上下文</td></tr>
<tr><td class="model-name" style="color:var(--overseas)">GPT-6 Sol / Luna</td><td>9/23 · 近 Astra 训练 · 更省</td><td class="good">ChatGPT 日常付费默认</td><td class="bad">极限仍选 Astra</td></tr>
<tr><td class="model-name" style="color:var(--overseas)">Claude Fable 5.1</td><td>$10/$50 · 缓存读 $0.25 · Arena #1</td><td class="good">长程 Agent 最强公开档</td><td class="bad">日常请先用 Opus 5.5</td></tr>
<tr><td class="model-name" style="color:var(--overseas)">Claude Opus 5.5</td><td>成本↓~40% · 速度↑~30% · 近 Fable</td><td class="good">日常长程 Coding 主力</td><td class="bad">中文、梯子；刊例看官方</td></tr>
<tr><td class="model-name" style="color:var(--overseas)">Gemini 3.8 Flash</td><td>$0.75/$3.75 至 12/31 · 元旦翻倍</td><td class="good">速度、工程 Agent</td><td class="bad">不是 3.1 Pro 替代</td></tr>
<tr><td class="model-name" style="color:var(--overseas)">Grok 4.7</td><td>$2/$6 · ~2.1T · 长任务</td><td class="good">工程编码、长跑 Agent</td><td class="bad">逊于 Fable / Astra</td></tr>
<tr><td class="model-name" style="color:var(--overseas)"><a href="decision-model-plan.html">Jev / Laya</a></td><td>决策专用 · 云 API / 开源权重</td><td class="good">软件内结构化决策</td><td class="bad">不聊天、不生成文本</td></tr>
</tbody>
</table>
</div>
</div>
<div class="cp-section-hd"><span class="cp-flag">🏆</span> Arena Agent 榜 Top 10（2026-09-05）</div>
<p style="font-size:.82rem;color:var(--muted);margin:0 0 .6rem;">按真实 Agent 会话的净提升排序，不是 SWE-bench。来源 <a href="https://arena.ai/leaderboard/agent" target="_blank" rel="noopener noreferrer">arena.ai/leaderboard/agent</a> · 228 万会话 / 59 个模型。刊例为百万 token 输入/输出。</p>
<div class="cp-matrix-wrap">
<table class="cp-matrix">
<thead>
<tr><th>#</th><th>模型</th><th>净提升</th><th>确认成功</th><th>任务成本 P50</th><th>刊例 $/M</th></tr>
</thead>
<tbody>
<tr><td>1</td><td class="model-name" style="color:var(--overseas)">Claude Fable 5.1 (Max)</td><td>15.87%</td><td>22.45%</td><td>$4.32</td><td>$10 / $50</td></tr>
<tr><td>2</td><td class="model-name" style="color:var(--overseas)">Claude Opus 5 (High)</td><td>12.68%</td><td>13.30%</td><td>$2.43</td><td>$5 / $25</td></tr>
<tr><td>3</td><td class="model-name" style="color:var(--overseas)">Claude Opus 5 (Max)</td><td>11.67%</td><td>14.89%</td><td>$3.66</td><td>$5 / $25</td></tr>
<tr><td>4</td><td class="model-name" style="color:var(--overseas)">Claude Fable 5 (High)</td><td>10.23%</td><td>7.37%</td><td>$2.37</td><td>$10 / $50</td></tr>
<tr><td>5</td><td class="model-name" style="color:var(--overseas)">GPT-5.6 Sol (xHigh)</td><td>9.31%</td><td>6.68%</td><td>$1.22</td><td>$4 / $20</td></tr>
<tr><td>6</td><td class="model-name" style="color:var(--overseas)">Claude Opus 4.8 (High)</td><td>9.17%</td><td>5.27%</td><td>$1.56</td><td>$5 / $25</td></tr>
<tr><td>7</td><td class="model-name">Kimi K3 (Max)</td><td>7.96%</td><td>16.58%</td><td>$0.82</td><td>$3 / $15</td></tr>
<tr><td>8</td><td class="model-name" style="color:var(--overseas)">Claude Sonnet 5 (High)</td><td>7.41%</td><td>2.80%</td><td>$0.87</td><td>$2 / $10</td></tr>
<tr><td>9</td><td class="model-name" style="color:var(--overseas)">GPT-5.5 (xHigh)</td><td>7.21%</td><td>2.07%</td><td>$1.07</td><td>$5 / $30</td></tr>
<tr><td>10</td><td class="model-name">Tencent Hy4 preview</td><td>6.92%</td><td>13.56%</td><td>$0.27</td><td>$0.83 / $2.50</td></tr>
</tbody>
</table>
</div>
<p style="font-size:.76rem;color:var(--muted);margin:.5rem 0 0;">国内开源侧：Hy4 #10、GLM-5.2 Max #12、Qwen 3.8 Max #16、DeepSeek-V4-Pro High #18、GLM-5.3 Max #22。海外速度档 Gemini 3.8 Flash High #17（3.7 已落到 #35）。</p>
<div class="cp-section-hd"><span class="cp-flag">🎯</span> 场景选型推荐</div>
<div class="cp-scenario">
<div class="cp-scenario-head">📚 场景 1：读 100 万字论文/法规</div>
<div class="cp-scenario-body">
<span class="candidate">候选：DeepSeek-V4.1-Flash、Qwen 3.8 Max、Gemini 3.1 Pro</span><br>
<span class="recommend">👉 推荐：日常快调用 DeepSeek-V4.1-Flash（<code>deepseek-flash</code>）；要稳长程先看 GLM-5.3</span>
<ul class="reason">
<li>1M 上下文 + CSA/HCA，长上下文推理 FLOPs 约为 V3.2 的 27%</li>
<li>Engram 条件记忆可 O(1) 检索超长文档条款</li>
</ul>
</div>
</div>
<div class="cp-scenario">
<div class="cp-scenario-head">💻 场景 2：AI 写完整 Web 应用（前端+后端+部署）</div>
<div class="cp-scenario-body">
<span class="candidate">候选：GLM-5.3、Kimi K3、Claude Opus 5.5、<strong>Grok 4.7</strong></span><br>
<span class="recommend">👉 推荐：GLM-5.3（国内）/ Claude Opus 5.5 或 Grok 4.7（海外）</span>
<ul class="reason">
<li>GLM-5.3 已推到全部 Coding Plan；高峰积分全额，其余时段五折</li>
<li>Grok 4.7（2026-09-21/22）长任务 coding/agent；API 报道约 $2/$6</li>
<li>日常稳妥用 Opus 5.5；冲极限用 Fable 5.1（Arena Agent #1，缓存更便宜）</li>
</ul>
</div>
</div>
<div class="cp-scenario">
<div class="cp-scenario-head">⚡ 场景 3：7×24 客服系统</div>
<div class="cp-scenario-body">
<span class="candidate">候选：MiniMax M2.7、Gemini 3.8 Flash、Gemma 4</span><br>
<span class="recommend">👉 推荐：MiniMax M2.7（云端）/ Gemma 4（端侧）</span>
<ul class="reason">
<li>M2.7 约 30 tok/s，10B 激活参数适合高并发</li>
<li>端侧隐私场景选 Gemma 4 E2B</li>
</ul>
</div>
</div>
<div class="cp-scenario">
<div class="cp-scenario-head">🤖 场景 4：开源 Agent 底座（Hermes + 国产 API）</div>
<div class="cp-scenario-body">
<span class="candidate">候选：DeepSeek-V4-Pro、GLM-5.3、Qwen 3.8 Max、Hy4 preview</span><br>
<span class="recommend">👉 推荐：GLM-5.3 + Hermes Agent；要开源低价看 Hy4</span>
<ul class="reason">
<li>社区主流：Hermes 编排 + GLM-5.3 长程 Coding</li>
<li>Hy4 preview（8/28）Apache 2.0，Arena Agent #10，API 约 $0.83/$2.50</li>
<li>要开源 Max 档智商用 Qwen 3.8 Max；本机 24GB 用 <a href="local-model-plan.html">Qwen3.8-27B</a></li>
</ul>
</div>
</div>
<div class="cp-scenario">
<div class="cp-scenario-head">🛡️ 场景 5：长程 Agent（Arena 2026-09-05）</div>
<div class="cp-scenario-body">
<span class="candidate">候选：Claude Fable 5.1、Claude Opus 5.5、Kimi K3、Hy4 preview</span><br>
<span class="recommend">👉 推荐：冲极限 Fable 5.1 / 日常 Opus 5.5；国内 Kimi K3 或 Hy4</span>
<ul class="reason">
<li>Fable 5.1（9/1）接档 Fable 5，Arena Agent 总分第一；缓存读降到 $0.25</li>
<li>日常主力用 Opus 5.5；极限定用 Fable 5.1（Arena Agent #1）</li>
<li>国内：Kimi K3 Max 第 7；Hy4 preview 第 10（preview，单价更低）</li>
</ul>
</div>
</div>
<div class="cp-summary">
<h2>📌 一句话速查</h2>
<div class="cp-summary-grid">
<div class="cp-summary-item"><span class="cp-badge badge-deepseek">快调用</span> DeepSeek <span class="arrow">→</span> V4.1-Flash（<code>deepseek-flash</code>）</div>
<div class="cp-summary-item"><span class="cp-badge badge-glm">长程 Coding</span> 工程交付 <span class="arrow">→</span> GLM-5.3</div>
<div class="cp-summary-item"><span class="cp-badge badge-kimi">Agent</span> 多步工作流 <span class="arrow">→</span> Kimi K3 / Hy4</div>
<div class="cp-summary-item"><span class="cp-badge badge-qwen">开源旗舰</span> Max 档 <span class="arrow">→</span> Qwen 3.8 Max</div>
<div class="cp-summary-item"><span class="cp-badge badge-minimax">速度</span> 实时交互 <span class="arrow">→</span> MiniMax M2.7</div>
<div class="cp-summary-item"><span class="cp-badge badge-gemma">端侧</span> 离线/隐私 <span class="arrow">→</span> Gemma 4 / <a href="edge-model-plan.html">端侧模型</a></div>
<div class="cp-summary-item"><span class="cp-badge badge-gpt">通用</span> ChatGPT <span class="arrow">→</span> GPT-6 Sol</div>
<div class="cp-summary-item"><span class="cp-badge badge-claude">深度</span> 日常 Coding <span class="arrow">→</span> Opus 5.5</div>
<div class="cp-summary-item"><span class="cp-badge badge-claude">长程</span> Agent 极限 <span class="arrow">→</span> Fable 5.1</div>
<div class="cp-summary-item"><span class="cp-badge badge-gemini">速度档</span> Google <span class="arrow">→</span> Gemini 3.8 Flash</div>
<div class="cp-summary-item"><span class="cp-badge badge-overseas">Coding/Agent</span> 工程编码 <span class="arrow">→</span> Grok 4.7 / Claude Opus 5.5</div>
</div>
<p style="margin-top:1rem;font-size:0.88rem;color:var(--muted);text-align:center;">没有「最好的模型」，只有「最适合你场景的模型」。Agent 名次来自 <a href="https://arena.ai/leaderboard/agent" target="_blank" rel="noopener noreferrer">Arena Agent</a>（2026-09-05，228 万会话）。</p>
</div>
<div class="cp-footer">
<p>数据来源：各模型官方技术博客与论文 · Agent 名次对照 <a href="https://arena.ai/leaderboard/agent" target="_blank" rel="noopener noreferrer">Arena Agent Leaderboard</a>（2026-09-05）</p>
<p><a href="index.html">← 返回 AINav 首页</a> · <a href="model-finder.html">模型推荐器</a> · <a href="local-model-plan.html">本地模型</a> · <a href="edge-model-plan.html">端侧模型</a> · <a href="decision-model-plan.html">决策模型</a> · <a href="low-filter-api-plan.html">低过滤云 API</a> · <a href="free-tier.html">🆓 免费额度</a> · <a href="token-optimization.html">Token优化</a> · <a href="coding-plan.html">编程套餐横评</a> · <a href="agent-plan.html">Agent横评</a></p>
</div>
</div>
<script>
(function(){
var btns=document.querySelectorAll('.theme-btn');
var html=document.documentElement;
function apply(t){html.setAttribute('data-theme',t);localStorage.setItem('cp-theme',t);btns.forEach(function(b){b.classList.toggle('is-active',b.getAttribute('data-theme-set')===t)})}
var saved=localStorage.getItem('cp-theme')||'system';apply(saved);
btns.forEach(function(b){b.addEventListener('click',function(){apply(b.getAttribute('data-theme-set'))})});
})();
</script>
<script>document.addEventListener("click",function(e){var c=e.target.closest(".plan-promo-code");if(!c)return;e.preventDefault();var code=c.getAttribute("data-code")||c.textContent.trim();if(navigator.clipboard&&navigator.clipboard.writeText){navigator.clipboard.writeText(code).then(function(){var o=c.textContent;c.textContent=code+" 已复制";setTimeout(function(){c.textContent=o},1500)})}});</script>
<script src="plan-nav.js" data-active="model-plan.html"></script>
<script src="plan-meta-i18n.js"></script>
<script src="thinking-framework.js"></script>
<script src="./back-to-top.js"></script>
</body>
</html>