Kimi K3.1 Pro 三次全过率仅 45%，落后 GPT-6 Astra 的 63%
ArtificialAnlys · x · 2026-11-07
Artificial Analysis notes that solving a task once doesn't mean consistent performance — consistency is a key frontier-model advantage on VIALS-AA. Across 161 life-sciences visual reasoning tasks attempted three times each, Kimi K3.1 Pro answered 119 of 161 correctly at least once (GPT-6 Astra: 118), but requiring all three attempts to pass flips the picture: Kimi passes 73 tasks (45%) while Astra passes 101 (63%).
The takeaway: leaderboards based on "passed at least once" substantially overstate weaker models — the stability gap is far larger than the single-attempt score gap.
所属事件:Artificial Analysis 发布生命科学视觉推理榜 VIALS-AA(5 条相关)→
「模型」频道最新
- Meta 推出 Muse Wonder 专用模型:遗忘减少 2.4 倍,卡壳改善 67% — AIatMeta · 2026-11-07
- 165 个生成世界实测:局部交互 83 分,有状态流程仅 30 分 — nptacek · 2026-11-07
- 激活转向实验:单个神经元即可让 Qwen3-30B 换「人格」,约1%回复见效 — Sauers_ · 2026-11-07
- OpenAI 与 Anthropic 互相观望按兵不发，大模型发布陷入僵局 — kimmonismus · 2026-11-07
- Deep Learning Weekly #480:提「Gemini 4 Carbon」与模型路由论文 — dl_weekly · 2026-11-07
- PostTrainBench：AI 智能体后训练得分半年从 23% 升至 45% — karinanguyen · 2026-11-07