Meta 的 Muse Spark Open 成为最强美国开源模型,加上 Reflection 的 Beam,美国开源权重开始回追中国模型的前沿差距。
2026-10-18
— 开源模型卷到 552B 参数,安全事件却提醒我们:地基还没稳。
DeepSeek 发布 V4.1-Flash-1017,在推理、编码和 agent 基准上全面提升,同时缩短输出并降低成本,成为开源权重模型新标杆。OpenCode 遭遇安全入侵,用户凭据泄露,官方敦促轮换密钥。Perplexity 开源 SPACE 沙箱,为 AI agent 提供基于 Firecracker microVM 的生产级隔离。StepFun 与 Meta 等也在开源权重领域持续加码,美国开源模型开始回追前沿。
头条
DeepSeek V4.1-Flash-1017 发布:552B MoE 开源模型全面刷新基准多源事件 ×6
DeepSeek 于 10 月 17 日发布 V4.1-Flash-1017,这是基于 Causal Encoder-Decoder (CED) 架构的稀疏 MoE 模型,552B 总参数、预填充激活 8B、解码激活 16B,支持 1M 上下文和原生图像理解。相比 9 月版本,DeepSWE v1.1 从 74.2 升至 77.0,Terminal-Bench 4.0 从 31.2 升至 42.3,同时任务执行时间和输出长度均下降。 为什么重要:该模型以 MIT 许可开放权重,并通过 deepseek-flash API 提供服务,同时开放 Search 和 Fetch API,为 coding agent 和长时任务提供了性能/成本帕累托前沿的新选项,直接冲击闭源模型在 agent 场景的定价逻辑。
评论区普遍认可其性能强、性价比高,但也有人质疑基准成绩存疑,实际表现未必胜过 Sol 等闭源模型。
OpenCode 遭入侵,用户凭据泄露,官方敦促轮换密钥
OpenCode 官方确认于 10 月 17 日夜间发生安全入侵,涉及大规模蒸馏攻击,导致用户凭据泄露。官方已发布公告,建议所有用户立即轮换 API 密钥。 为什么重要:OpenCode 是面向开发者的 AI 编程工具,凭据泄露意味着攻击者可能直接调用用户的云服务 API,造成资损或数据泄露。对依赖该工具接入 LLM 的工程师而言,这是一次供应链安全的现实警示。
评论普遍认为 OpenCode Console 泄露用户 API 密钥早有预警且团队难辞其咎,但也有人认为仅影响 Console 服务、本地 CLI 用户不受波及。
Perplexity 开源 SPACE:基于 Firecracker microVM 的 AI Agent 沙箱
Perplexity 开源了 SPACE,这是支撑 Perplexity Computer 每个任务的生产级沙箱平台。每个 agent 任务运行在独立的 Firecracker microVM 中,凭据和出站流量从 VM 外部控制,代码以 Apache 2.0 许可发布在 GitHub。 为什么重要:对于构建 coding 或 browsing agent 的团队,SPACE 提供了一个经过生产验证的自托管隔离层。其 0/108 的 VM 逃逸拦截记录和 60ms 的中位创建延迟,为 agent 安全执行提供了可直接复用的基础设施参考。
StepFun 发布 Step 5 Preview 开放权重:600B MoE、27B 激活、1M 上下文
StepFun 发布 Step 5 Preview 开放权重模型,总参数 600B、每 token 激活 27B,支持最高 1M token 上下文。FP8 和 BF16 检查点已在 Hugging Face 和 ModelScope 上线,面向长时 agentic 工作负载。 为什么重要:这是又一个面向 coding agent 和文档分析的自托管选项。虽然 Terminal-Bench 4.0 仅 33.3%,但在 AA-LCR 长上下文推理上达 88.3%,且硬件门槛明确(FP8 需 4x B300 或 8x H200),为有数据中心 GPU 资源的团队提供了新的部署选择。
每天早晨,一份为你精选的科技日报
网页看大盘,订阅拿专属:AI 按你的兴趣为你精选、可汇入你的私有 RSS,附社区观点——每天早晨直达邮箱,永久免费。
已发布 98 期 · 每天筛过 150+ 条只留值得读的 30 条
AI 动态
Replit CEO 宣布将默认模型从 Anthropic 切换为 OpenAI,理由是成本、合作与内部基准表现。
稀疏自编码器(SAE)特征的行为可组合性研究:特征部分可加,但准确预测和 steering 需要交互项。
🤖Sparse autoencoder features are partially compositional but require interaction terms for accurate behavioral predictions and steering.
Virtual Head Attention (VHA) 通过轻量线性变换同时降低 KV cache、训练/prefill FLOPs 并保持模型质量。
🤖Virtual Head Attention improves long-context model efficiency by reducing KV cache and compute while preserving quality through lightweight linear query and output transformations.
EDEN 方法在 RLVR 的 rollout 阶段注入锚定嵌入噪声,提升轨迹多样性并强化组相对优势信号。
🤖EDEN improves reinforcement learning with verifiable rewards by injecting anchored embedding noise into rollout conditioning to increase diversity and strengthen group-relative advantage signals without altering inference.
开发与开源
Pele 将 Mermaid 语法渲染为可主题化 SVG,gzip 后仅 230 kB、零依赖,渲染速度比 Mermaid.js 快两个数量级。
esp32sim 在浏览器中实现 ESP32 指令级仿真,可启动真实 mask ROM 和未修改固件,支持 ESP32-S3/C3/C6。
llmspeed.dev 让用户在购买硬件前预估本地 LLM 的 token 生成速度,社区认可想法但认为估算精度和配置选项有限。
评论普遍认可该工具帮助预估本地LLM硬件性能的想法,但也有人认为其速度估算不准、模型和硬件配置选项有限。
通过自适应 KV streaming 和 RDNA 4/ROCm 优化,Qwen 27B 在 16GB RX 9070 XT 上以 80+ t/s 解码跑满 262k 上下文。
社区热议
用 2 台游戏 PC + M4 Pro MacBook + 2 台 iPhone 17 Pro Max 通过以太网/USB-C 分布式运行 Qwen3.8 Flash Next,60k 上下文下 37-39 tok/s。
讨论 Qwen 3.8 Flash Next 的 n-gram 表(PLE)在常见 GGUF 中被过度量化,影响性能,建议使用完整表。
一位基础设施工程师首次接触 Kubernetes 的反思文章引发共鸣,评论区集中批评 Helm 字符串模板和 YAML 嵌套复杂度。
评论普遍认可文章可读性,并集中批评 Helm 字符串模板和 YAML 层层嵌套,但也有人认为问题不在 YAML 或模板层,K8s 作为期望状态协调引擎的价值不应因此被否定。
开发者分享从怀疑到接受本地 AI 的经历,感叹 llama.cpp 在过去一年多的进步。
GitHub Trending
Star anthropics / claude-plugins-official Official, Anthropic-managed directory of high quality Claude Code Plugins.
Sponsor Star MadsLorentzen / ai-job-search The job search that runs on your machine. AI job application framework built on Claude Code: evaluate postings, tailor CVs, write cover letters, prep interviews. Fork it and own it.
Sponsor Star affaan-m / ECC The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
Star HKUDS / CLI-Anything "CLI-Anything: Making ALL Software Agent-Native" -- CLI-Hub: https://clianything.cc/
Star K-Dense-AI / scientific-agent-skills Turn any AI agent into an AI Scientist. The #1 Agent Skills library for science, used by 250,000+ scientists worldwide. 177 ready-to-use validated skills plus 100+ scientific databases covering biology, chemistry, medicine, and drug discovery. Compatible with Cursor, Claude Code, Codex, Pi, Antigravity, and the open Agent Skills standard.
Star akitaonrails / ai-memory Solution for long term memory for agent coding CLIs and to facilitate handoff between different agent vendors
Star stablyai / orca Orca is the place for working with a fleet of agents (e.g. Grok Bot, Claude Code). Run any coding agent with your own subscription. Available on desktop, mobile and remote runtime.
Star anthropics / claude-code Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.
Sponsor Star debpalash / VoiceStudio VoiceStudio is the open-source, fully-local ElevenLabs alternative — voice cloning, voice design, video dubbing, dictation, transcription & audiobook creation in 646 languages.
更多值得一看(内容池 18 条)
Listen to the session or watch below Can AI design new life forms? In 2025, Stanford University PhD student Samuel King came up with a preliminary answer when he used a generative AI model to propose genetic blueprints for microscopic viruses. It isn’t yet an example of AI-generated life, but that could be next. Watch…
I’ve been building VeloxDB, an open-source database client built with Rust and Tauri, and it just crossed 650 GitHub stars. The idea came from wanting a database tool that feels fast, lightweight, and comfortable to use without sending everything through a cloud service. VeloxDB supports PostgreSQL, MySQL, SQLite, and more. It runs locally, connects directly to your databases, and includes a visual data explorer and query editor. It’s been a fun journey building this in the open, learning from f
A Meta spokesperson insisted that the updated terms did not reflect a change in the way Meta crawls the web.
After backlash, Meta is updating its terms again to remove language that suggested it could scrape content from sites, including news sites, to train its AI. Specifically, the following language no longer appears on the commercial terms page: … Meta may use any content from such domains, web pages, or URLs to provide, develop, improve, […]
The DNA inside our cells isn’t just a series of instructions, but something far more dynamic. Like scientists, AI models don't yet have the data to properly decode it.
Article URL: Comments URL: Points: 38 # Comments: 6
As video models enable longer and more realistic interactive experiences, maintaining a world consistent with its history becomes essential. This requires both preserving previously observed content and ensuring coherent world evolution over time. World memory is a critical capability to evaluate alongside visual fidelity and video duration. We introduce WorldMemBench, a benchmark for systematically evaluating world memory in video world models. Across 260 test cases, it assesses static memory t
Hello! We're two friends working in tech, and we've been building Safebucket: a simple, S3-agnostic, open-source alternative to WeTransfer, Dropbox and Lockself. What's new in v0.8.0: Customisable links: Create a quick share link with a custom slug File explorer improvements: Bulk actions (move, download and trash) Move files over with drag and drop UI/UX redesign We're open to feedback, contributions, and questions. What features would you like to see next? Github: Documentation:
I understand this kind of product may not be exactly right for this subreddit, but I still think it deserves some attention, as it's an LLM frontend for local engines in the first place, and can be used for general-purpose chats with LLMs once lore extraction is disabled in the settings. Otaku is built with a focus on three key ideas: maintain writing quality as the story grows; keep the context small , even for long stories, without losing the details that matter; offer high-quality interfaces