Dongxi 东锡 NLP· @dongxi_nlp · X·· 2026-08-08AI 评分52
AI 导读
DeepSeek V4 Flash 在 ARC-AGI-1 上以低推理预算即得约 84 分，最高约 89 分，额外推理仅多解 5 题，表明其后训练产生了异常高效的抽象推理。该模型 ARC-AGI-2 得分 61.4%（$0.04/任务），ARC-AGI-1 得分 89.0%（$0.02/任务），树立了性价比帕累托前沿新标准。
正文 · 原文
DeepSeek reaches close to its best score even when given a smaller reasoning budget.
ARC‑AGI‑1:
Low solves about 84.
Max solves about 89.
Extra reasoning recovers only five additional puzzles.
This suggests DeepSeek’s post-training produces unusually efficient abstract reasoning.
DeepSeek V4 Flash from @deepseek_ai on ARC-AGI (Verified): - ARC-AGI-2: 61.4%, $0.04/task - ARC-AGI-1: 89.0%, $0.02/task DeepSeek V4 Flash sets the new standard on the cost-to-performance Pareto frontier.在 X 查看被引用的帖子
来源：Dongxi 东锡 NLP · x.com