Shanghai AI4S open-sources REMORY, a 1.9B memory module for recursive self-improvement
The Shanghai Academy of AI for Science, with Fudan University and Shanghai Chuangzhi Institute, open-sourced REMORY, a 1.9B-parameter memory module that enables small models to run independently for days via recursive self-improvement, achieving results near GPT-6.1-Terra. Trained fully self-supervised on two H800 GPUs, REMORY leads several key benchmarks and is available on arXiv, Hugging Face, GitHub, and the Xinghe Qizhi platform.
Coverage timeline
机器之心机器之心
继 Apex-35B-V1.0 凭借领先 Flint 的表现引发关注之后，开源 RSI 模型又迎来一份亮眼成绩单： REMORY 仅凭一个 1.9B 的记忆模块，用两张 H800 实现完全自监督训练，在多项关键评测中领跑；驱动小模型持续数日递归改进，产出接近 GPT-6.1-Terra 。 本文第一作者夏翰宸为上海科学智能研究院研究员，研究方向为 auto research 和智能体记忆。本文的通讯作者朱思语是复旦大学教授、上海科学智能研究院 AI 科学家、上海创智学院全时导师。 论文链接：https://arxiv.org/pdf/2610.11287 Huggingface主页：https://huggingface.co/papers/2610.11287 Github 主页：https://github.com/fudan-generative-vision/Remory 星河启智平台主页：https://aistudio.ai4s.com.cn/partner/research-plaza/galaxy-model/model/Qwen3.8-27B-REMORY-1.9B RSI（Recursive Self-Improvement）已经成为整个 AI 行业押注的方向。Recursive Super Intelligence 公司由前 Salesforce 首席科学家 Richard Socher 领衔，联创包括前 Meta 明星研究员田渊栋以及来自 OpenAI、Google DeepMind 等机构的顶尖人才，在 10 月 22 日发布了 Flint（40B）；随后 10 月 28 日，由清华大学人工智能学院助理教授陈超勇创立的超衍智能，发布了其首个旗舰模型 Apex-35B-V1.0， 投身 RSI 创业。在这一聚集了硅谷多方押注的新星创企、清华最年轻助理教授团队等选手的大热方向，近期，由上海科学智能研究院（简称上智院）携手上海创智学院、复旦大学组成的 REMORY 团队， 在核心开发仅两人的情况下，用 1.9B 可训练参数，和两张 H800，在多个高难度评测中交出了领先成绩 ： 越过 /compact，让经验持续生效 你日常使用的大模型的上下文窗口是多少？2M？1M？还是 0.2M？一个致力于完成超长任务或和用户长期交互的 agent，上下文压缩
Related stories
MemTensor Releases MemT 1.0, Claiming First Memory-Native Open-Source Foundation Models
On October 20, 2026, Memory Tensor released MemT 1.0, a family of memory-native foundation models that encode historical information into continuously updated internal memory states, enabling parameterized memory to participate in subsequent generation. The release includes MemT-1.0-Flash (288B parameters, 17B active, 336M dynamic memory) and MemT-1.0-Nano (36B parameters, 3B active, 10M dynamic memory), with open-source checkpoints and a technical report, marking the first engineering validation of memory-native models at near-300B parameter scale. The approach integrates a parallel memory computation path alongside existing attention in selected Transformer modules, making memory a learnable capability of the model itself.
2 reports · 2 sources
OpenAI releases GPT-6.1 Terra with computer-use focus, API price halved
OpenAI announced GPT-6.1 Terra, a fast, efficient model emphasizing visual reasoning and computer use, priced at $1 per million input tokens and $5 per million output tokens. The standard API price is 50% lower than GPT-5.6 Terra and GPT-6.1 Sol, and the model is available in ChatGPT Work, Codex, and via API (gpt-6.1-terra) for Plus, Pro, Enterprise, and Edu users. In the Agents' Last Exam benchmark, Terra scored 15 points higher than GPT-5.6 Terra at medium reasoning intensity, with about one-third the cost, and matched GPT-6 Astra's performance at one-tenth the token price.
PKU Researchers Propose MapFlow to Reuse Attention Maps Across LLMs, Cutting Prefill Latency 15-20%
Researchers at Peking University's PKU-DAIR lab propose MapFlow, a method that reuses attention maps as intermediate states across different LLMs to avoid recomputing context when switching models. In evaluations, MapFlow reduces prefill latency by 15-20% and boosts throughput by up to 25% under a P90 first-token latency constraint in multi-turn routing scenarios, while maintaining generation quality. The paper has been accepted at SIGMOD 2027 and the code is open-sourced.
1 reports · 1 sources
Meta and academic researchers propose World-T³, a video world model with test-time training for long-term memory
Researchers from Meta Reality Labs, the University of Illinois Chicago, and Brown University propose World-T³, a camera-controllable video world model that uses test-time training to store historical observations in fast weights, keyed by camera pose, enabling consistent scene recall when revisiting locations. On a long-term revisit benchmark of 120 trajectories, it improves revisit PSNR from 13.52 dB to 18.31 dB and reduces absolute rotation error from 3.344° to 1.075°, while a 4-step distilled model runs at 20.6 FPS on a single NVIDIA B200 GPU at 1280×704 resolution.
News updates every hour. Subscribe in your RSS reader to get new stories automatically.
Subscribe