Weekly review
Weekly review · Oct 5 – Oct 11
Signals published Oct 5 – Oct 11 (Monday to Sunday), by priority. Derived entirely from published signals: no editor's text, no AI.
Signal titles are in English where the source is; the editors' weekly summary is written in Chinese.
Immediate attention
1 signalThis week's signals marked critical, or important and published within the last 30 days.
何恺明团队 VISTA：让模型回看原始画面，打满 ARC-AGI-3
MIT 何恺明团队的 VISTA 是一套视觉原生的智能体外壳：环境返回的每一帧都原样存档，模型推理时可以按需调回、放大、并排对比，基础模型不用重新训练。配上 Claude Opus 5.0，ARC-AGI-3 的 25 个公开游戏全部通关，相对人类动作效率得满分 100，比人类首次游玩的基线少用 57.4% 的动作。
Worth tracking
7 signalsThis week's changes worth keeping an eye on while they are confirmed.
MiMo-V2.6 技术报告：小米花 260 万美元跑一次强化学习
小米 MiMo 团队 10 月 8 日发布 MiMo-V2.6 技术报告：Pro（1.02T 总参数、激活 42B）与 Flash（310B、激活 15B）两款 MoE 模型，强化学习同时扩大批量、任务环境与评分算力。每步约 2.5 万条轨迹、27 亿至 37 亿 token，Pro 的强化学习花费约 260 万美元，其中评分器占 12.7%。训练全程冻结 MoE 路由，作弊轨迹控制在 2% 以下；训练动态、强化学习环境与框架开源。
DeepSeek-V4 的周期性盲点：信息换个位置，检索准确率差 40 个百分点
字节跳动 Seed 发现，采用分块 KV 缓存压缩的模型，检索准确率会随信息相对压缩窗口边界的位置周期性起伏。128K 上下文里，DeepSeek-V4-Flash-Base 在不同位置之间最多差 40.2 个百分点，后训练过的 V4.1-Flash 仍差 6.1 个百分点。起伏的周期正好等于压缩步长：V4 是 4 个 token，V4.1 是 2 个。平均分会把这种弱点藏起来。
Agent Lightning v1.0: A 3,500-Line Lightweight Agentic RL Framework for Training Agents with Real Harnesses
Training AI agents with reinforcement learning can be challenging because their tools, context, and decision-making are managed by complex frameworks. Agent Lightning connects existing agents to RL training, making it easier to improve them without rebuilding…
Claude Haiku 5.5
As previously promised, here's Anthropic's new fast, low cost model: Introducing Claude Haiku 5.5. The previous Haiku, 4.5, was very much showing its age. It came out almost a year ago, and was priced at $1/million input and $5/million output - relatively exp…
One Model Family, Two Gold-Level Results: Fine-Tuning Nemotron for IOI and IMO
One Model Family, Two Gold-Level Results: Fine-Tuning Nemotron for IOI and IMO
EmbeddingGemma 2: an open, lightweight multimodal embedding model
EmbeddingGemma 2: an open, lightweight multimodal embedding model
Sharing AI progress in mathematics
OpenAI publishes new results on open problems in mathematics from an internal frontier model and shares Lean proof formalizations and research details on GitHub.