{
 "generated": "2026-11-02",
 "node": {
  "id": "event:sig-0081",
  "type": "Event",
  "name": "Cognition SWE-2 在 Terminal-Bench 2.1 拿 92.8%，超过 Fable 5.1（91.4）与 GPT-6 Astra（89.9），基座是 Kimi K3",
  "slug": "sig-0081",
  "kind": "signal",
  "kind_zh": "行业信号",
  "date": "2026-09-10",
  "summary": "Cognition 09-10 博客：SWE-2 由 2.8T 参数的 Kimi K3 后训练而来，首次把 RL 扩到万亿参数量级，用单次 RL 同时训练所有推理强度档位。FrontierCode 1.1 Main 50.0%（Fable 5.1 为 50.9%、Astra 53.3%）但便宜 64%；Terminal-Bench 4 仍只有 27.3%，远低于 Astra 57.9%——旧版 Terminal-Bench 2.1 已经饱和，新版 4.0 才拉得开差距。tbench.ai 榜单页为客户端渲染，本次未能直接核对榜单本身。",
  "url": "https://cognition.ai/blog/swe-2",
  "evidence": [
   {
    "kind": "own",
    "label": "本站页面",
    "url": "https://xiaopingfeng.com/graph/signals/#sig-0081",
    "text": "AI 行业信号流（Benchmark）：Cognition SWE-2 在 Terminal-Bench 2.1 拿 92.8%，超过 Fable 5.1（91.4）与 GPT-6 Astra（89.9），基座是 Kimi K3"
   },
   {
    "kind": "other",
    "url": "https://cognition.ai/blog/swe-2",
    "text": "信号的原文出处（Terminal-Bench（新增源，尚未进 sources.md；经 Cognition 博客））"
   }
  ],
  "entry": "derived",
  "category": "benchmark",
  "category_zh": "Benchmark",
  "source_name": "Terminal-Bench（新增源，尚未进 sources.md；经 Cognition 博客）",
  "source_id": "bench-terminalbench",
  "degree": 0
 },
 "page": "https://xiaopingfeng.com/graph/e/event/sig-0081/",
 "out": [
  {
   "type": "placed_in",
   "sentence": "Cognition SWE-2 在 Terminal-Bench 2.1 拿 92.8%，超过 Fable 5.1与 GPT-6 Astra，基座是 Kimi K3出现在 Landscape 的「Benchmark」",
   "direction": "out",
   "id": "layer:ai-industry-signals--1",
   "name": "Benchmark",
   "status": "derived",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "url": "https://xiaopingfeng.com/graph/signals/#sig-0081",
     "text": "AI 行业信号流（Benchmark）：Cognition SWE-2 在 Terminal-Bench 2.1 拿 92.8%，超过 Fable 5.1（91.4）与 GPT-6 Astra（89.9），基座是 Kimi K3"
    }
   ]
  },
  {
   "type": "involves",
   "sentence": "Cognition SWE-2 在 Terminal-Bench 2.1 拿 92.8%，超过 Fable 5.1与 GPT-6 Astra，基座是 Kimi K3涉及Terminal-Bench",
   "direction": "out",
   "id": "org:terminal-bench",
   "name": "Terminal-Bench",
   "role": "issuer",
   "note": "信源：Terminal-Bench（新增源，尚未进 sources.md；经 Cognition 博客）",
   "status": "derived",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "url": "https://xiaopingfeng.com/graph/signals/#sig-0081",
     "text": "AI 行业信号流（Benchmark）：Cognition SWE-2 在 Terminal-Bench 2.1 拿 92.8%，超过 Fable 5.1（91.4）与 GPT-6 Astra（89.9），基座是 Kimi K3"
    }
   ]
  },
  {
   "type": "mentions",
   "sentence": "Cognition SWE-2 在 Terminal-Bench 2.1 拿 92.8%，超过 Fable 5.1与 GPT-6 Astra，基座是 Kimi K3提到Kimi",
   "direction": "out",
   "id": "app:kimi",
   "name": "Kimi",
   "status": "derived",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "名称匹配：「Kimi」出现在信号的标题里",
     "url": "https://xiaopingfeng.com/graph/signals/#sig-0081"
    }
   ]
  },
  {
   "type": "mentions",
   "sentence": "Cognition SWE-2 在 Terminal-Bench 2.1 拿 92.8%，超过 Fable 5.1与 GPT-6 Astra，基座是 Kimi K3提到Cognition",
   "direction": "out",
   "id": "app:cognition",
   "name": "Cognition",
   "status": "derived",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "名称匹配：「Cognition」出现在信号的标题里",
     "url": "https://xiaopingfeng.com/graph/signals/#sig-0081"
    }
   ]
  },
  {
   "type": "mentions",
   "sentence": "Cognition SWE-2 在 Terminal-Bench 2.1 拿 92.8%，超过 Fable 5.1与 GPT-6 Astra，基座是 Kimi K3提到Moonshot AI",
   "direction": "out",
   "id": "org:moonshot",
   "name": "Moonshot AI（月之暗面）",
   "status": "derived",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "名称匹配：「Kimi」出现在信号的标题里",
     "url": "https://xiaopingfeng.com/graph/signals/#sig-0081"
    }
   ]
  },
  {
   "type": "mentions",
   "sentence": "Cognition SWE-2 在 Terminal-Bench 2.1 拿 92.8%，超过 Fable 5.1与 GPT-6 Astra，基座是 Kimi K3提到Cognition",
   "direction": "out",
   "id": "org:cognition",
   "name": "Cognition",
   "status": "derived",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "名称匹配：「Cognition」出现在信号的标题里",
     "url": "https://xiaopingfeng.com/graph/signals/#sig-0081"
    }
   ]
  }
 ],
 "in": []
}