MiniMax (official)@MiniMax_AI · 6月13日33

appreciate it @SambaNovaAI 🤝 looking forward to M3 on RDUs

译SambaNovaAI 祝贺 MiniMax 发布 M3 开源模型，并表示未来将在其 RDUs 上支持 M3。MiniMax 表示感谢并期待这一合作。

MiniMax (official)@MiniMax_AI · 6月13日69

Run M3 locally today with @UnslothAI

译MiniMax-M3 是一款拥有 428B（23B 激活）参数、1M 上下文的新开源模型，性能与 Gemini 3.1 Pro 相当。可在 138GB 内存/显存上运行动态 2-bit GGUF 版本，或 165GB 上运行 3-bit 版本。在 @UnslothAI 的帮助下，今天即可本地运行 M3。

Google AI@GoogleAI · 6月13日40

Here’s what launched this week: — Gemini 3.5 Live Translate our latest audio model for live speech-to-speech translation — @NotebookLM got a major upgrade including agentic capabilities in chat, more advanced reasoning, and a suite of new output formats — Project Genie from @GoogleLabs is now available to Google AI Ultra 5x subscribers globally — Notebooks in @GeminiApp are now available in the European Economic Area, United Kingdom, and Switzerland — DiffusionGemma, our newest experimental open @googlegemma model that explores text diffusion, an exceptionally fast approach to text generation

译Google AI 本周推出多项更新：Gemini 3.5 Live Translate 是用于实时语音到语音翻译的最新音频模型；NotebookLM 获重大升级，加入智能体对话能力、更高级推理及新输出格式；来自 GoogleLabs 的 Project Genie 向 Google AI Ultra 5x 订阅者全球开放；GeminiApp 中的 Notebooks 在 EEA、英国、瑞士上线；同时发布实验性开源模型 DiffusionGemma，探索文本扩散技术，实现极快文本生成。

MiniMax (official)@MiniMax_AI · 6月13日41

M3 is now live on @parasail_io 🚀

译MiniMax M3 已在 @parasail_io 上线。这是首个具备前沿编码与智能体能力的开源权重模型，拥有 1M 上下文窗口和原生多模态理解。联系我们可以启动专属端点进行测试！🚀

jason@jxnlco · 6月13日46

codex for open source! just granted about another huge batch including some that you might recognize! tensorflow/tensorflow n8n-io/n8n twbs/bootstrap github/gitignore ytdl-org/youtube-dl vercel/next.js 30-seconds/30-seconds-of-code kubernetes/kubernetes papers-we-love/papers-we-love angular/angular neovim/neovim microsoft/web-dev-for-beginners florinpop17/app-ideas bitcoin/bitcoin gin-gonic/gin microsoft/playwright laravel/laravel gothinkster/realworld spring-projects/spring-boot tensorflow/models apple/swift unclecode/crawl4ai tldr-pages/tldr snowpackjs/astro embedchain/embedchain vim/vim pingcap/tidb jonnyburger/remotion aspnet/aspnetcore seleniumhq/selenium jqlang/jq immutable-js/immutable-js anncwb/vue-vben-admin pynecone-io/pynecone martinvonz/jj serverless-stack/serverless-stack manojvivek/responsively-app trekhleb/homemade-machine-learning sipeed/picoclaw spicetify/spicetify-cli vueuse/vueuse guidance-ai/guidance nautechsystems/nautilus_trader hshoff/vx preservim/nerdtree officedev/office-ui-fabric-react carlospolop/peass-ng reduxjs/reselect adonisjs/adonis-framework rizinorg/cutter facebookresearch/llama-recipes stackexchange/dapper resendlabs/react-email tomav/docker-mailserver lichess-org/lila google/libphonenumber apache/incubator-brpc googlechrome/chrome-app-samples hwchase17/langchainjs fanux/sealos argoproj/argo argoproj/argo-workflows rjsf-team/react-jsonschema-form secureauthcorp/impacket scylladb/scylla uuidjs/uuid cayleygraph/cayley cesiumgs/cesium eclipse-vertx/vert.x pyodide/pyodide jetstack/cert-manager rileytestut/altstore sunnyyoung/wechattweak-macos pydanny/cookiecutter-django pandas-profiling/pandas-profiling espanso/espanso ansible-semaphore/semaphore k9mail/k-9 nock/nock dotnet/aspnetcore.docs selectize/selectize.js mozilla-mobile/firefox-ios wanghongenpin/network_proxy_flutter webpack-contrib/webpack-bundle-analyzer alicevision/meshroom actions/virtual-environments jxnl/instructor theramu/fay svprogresshud/svprogresshud lexikos/autohotkey_l lipis/flag-icon-css redpanda-data/redpanda vega/vega mrjbq7/ta-lib uber/ludwig keplergl/kepler.gl devicons/devicon crossplane/crossplane openaccess-ai-collective/axolotl go-shiori/shiori audiokit/audiokit pyroscope-io/pyroscope px4/px4-autopilot quickwit-oss/quickwit vuecomponent/ant-design-vue-pro divanteltd/vue-storefront k2-fsa/sherpa-onnx jantimon/html-webpack-plugin mockery/mockery automattic/node-canvas divio/django-cms containers/skopeo kubernetes/kompose lucia-auth/lucia microsoft/fluentui-system-icons triton-inference-server/server pressly/goose altair-viz/altair pwndbg/pwndbg maplibre/maplibre-gl-js webtorrent/webtorrent-desktop hackmdio/codimd

译Codex 为开源项目提供免费授权，最新一批包括 TensorFlow、Next.js、Kubernetes、Angular、Swift、Spring Boot、Playwright、Vim、Bitcoin、n8n、Bootstrap、酷狗（30-seconds-of-code）等大量知名开源仓库，列表涵盖机器学习、前端框架、基础设施、游戏开发、数据库等众多领域。具体授权范围和细则未在推文中说明。

🚨 AI News | TestingCatalog@testingcatalog · 6月12日57

KIMI AI🔥: A new open-source “Kimi K2.7 Code” model has been released on APIs and Huggingface! > Improved coding & agent performance over K2.6 > Reasoning efficiency > Long-horizon coding Testing time 👀

译KIMI AI🔥: 一个新的开源“Kimi K2.7 Code”模型已在 API 和 Huggingface 上发布！ > 相比 K2.6，编码与智能体性能提升 > 推理效率 > 长时域编码 测试时间 👀

X.PIN@thexpin · 6月12日72

Just attended Huawei Developer Conference 2026 in person. Huawei's "Doom Slayer," Richard Yu (Yu Chengdong), is back — this time taking over the LLM. He unveiled Huawei's new openPangu 2.0 model and was unusually candid about the shortcomings. Huawei short on its own compute. And Huawei badly needs AI talent: "We can't match internet companies on salary — the people building LLMs with us run on belief and conviction." At Huawei, when Yu gets handed a business, it's because the fight is brutally hard. And he keeps winning — phones from scratch in 2011, the Seres car partnership in 2021. In June 2025, Pangu was alleged to be highly similar to Qwen-2.5, and the responsible exec left. Yu taking over means Huawei's models are behind and need to catch up fast.

译华为开发者大会2026上，余承东正式接管华为LLM业务，发布openPangu 2.0模型。他坦诚华为自有算力不足，且AI人才招揽困难：“我们薪资拼不过互联网公司，一起做大模型的人靠信念和热情坚持。”2025年6月，旧版Pangu模型曾被指与Qwen-2.5高度相似，相关高管离职。余承东此前曾从零打造华为手机、推动赛力斯汽车合作。此次接管意味着华为模型落后，需加速追赶。

Chubby♨️@kimmonismus · 6月12日66

Moonshot just released Kimi-K2.7 code, a huge upgrade to Kimi-K2.6! Big jump over K2.6: +21.8% on Kimi Code Bench v2 +11.0% on Program Bench +31.5% on MLS Bench Lite It also uses 30% fewer reasoning tokens, follows instructions better, and improves long-horizon coding tasks. 6x High-Speed Mode is coming soon. Good to see open source competition catching up

译Moonshot 发布并开源 Kimi-K2.7-Code 编程模型，相比 K2.6 在多个基准上大幅提升：Kimi Code Bench v2 提高 21.8%，Program Bench 提高 11.0%，MLS Bench Lite 提高 31.5%。推理效率优化，推理 token 使用量降低 30%，指令遵循与长时编码任务成功率提升。即将推出 6 倍高速模式。模型现已通过 Kimi API 和 Kimi Code 开放使用。

OpenClaw🦞@openclaw · 6月12日58

OpenClaw 2026.6.6 🦞 🔒 Tighter security boundaries 💬 Safer Telegram + iMessage delivery 🧠 Claude Fable 5 + OpenRouter OAuth ⚡ Faster Control UI first replies Less weird, more work done. https://github.com/openclaw/openclaw/releases/tag/v2026.6.6

译OpenClaw 2026.6.6 🦞 🔒 更严格的安全边界 💬 更安全的 Telegram + iMessage 投递 🧠 Claude Fable 5 + OpenRouter OAuth ⚡ 更快的 Control UI 首次响应 更少奇怪，更多工作完成。 https://github.com/openclaw/openclaw/releases/tag/v2026.6.6

meng shao@shao__meng · 6月12日70

Kimi 开源发布最新编码模型「Kimi-K2.7-Code」，在 K2.6 基础上针对编程 Agent 做专项优化的版本，目标很明确：长链路编码任务的成功率更高，推理 token 更少！ # 三个核心改进 1. 编码：全面进步，尚未登顶 相对 K2.6，三项编码基准均有提升：Kimi Code Bench v2 +21.8%（50.9→62.0），Program Bench +11.0%，MLS Bench Lite +31.5%（涨幅最大，但绝对分仍低）。 与 GPT-5.5、Opus 4.8 比：综合编码任务差距明显缩小；MLS 与 GPT-5.5 基本持平；Program Bench 仍落后 GPT-5.5 一截。结论：稳健迭代，不是 leapfrog。 2. Agent：MCP 是亮点 Kimi Claw 24/7（长周期协作）和 MCP Atlas 均有提升，但仍落后于两大闭源模型。 MCP Mark Verified（81.1）超过 Opus 4.8（76.4） 是最有说服力的结果——覆盖 Notion、GitHub、Postgres、Playwright 等真实 MCP 环境，且经人工复核。说明 K2.7 在多工具编排上已具竞争力，GPT-5.5（92.9）仍是天花板。 3. 效率：更少 token，更高分 K2.7 不只提分，还降 reasoning token（官方称整体约 -30%）： · Kimi Code Bench v2：62k→48k token，分数 51%→62% · Program Bench：176k→102k token（-42%），分数 48%→53% · MLS Bench Lite：42k→38k token，分数 27%→35% 对 Agent 的实际意义：同样预算能跑更多步，长任务更省、更稳。 # 关键技术特性 1. 强制 Thinking 模式 不支持 Instant 模式；推荐 temperature=1.0、top_p=0.95。面向复杂推理，而非快速补全。 2. Preserve Thinking（强制开启） 多轮对话中保留完整 reasoning 内容，不可关闭。对编码 Agent 很重要——模型能引用先前推理链中的中间结论，减少上下文丢失。 3. Interleaved Thinking + Multi-Step Tool Call 与 K2 Thinking 相同设计：推理与工具调用交替进行，适合「想一步、调一步、再看结果」的 Agent 循环。 4. 多模态 支持图像和视频输入（官方 API 已支持；第三方 vLLM/SGLang 部署的视频能力仍为实验性）。 开源地址： https://huggingface.co/moonshotai/Kimi-K2.7-Code

译Kimi 开源发布最新编码模型 Kimi-K2.7-Code，基于 K2.6 优化。编码基准全面提升：Kimi Code Bench v2 提高 21.8%，Program Bench +11.0%，MLS Bench Lite +31.5%。推理 token 整体降低约 30%。Agent 方面，MCP Mark Verified 得分 81.1，超过 Opus 4.8（76.4），GPT-5.5（92.9）仍为天花板。技术特性：强制 Thinking 模式、Preserve Thinking、Interleaved Thinking+多步工具调用，支持图像和视频输入。可通过 Kimi API 和 Kimi Code 使用，6x 高速模式即将推出。开源地址：HuggingFace 上的 moonshotai/Kimi-K2.7-Code。

Kimi.ai@Kimi_Moonshot · 6月12日70

🌘 Kimi-K2.7-Code, our latest coding model, is now released and open-sourced! 🔷 Improved coding & agent performance over K2.6: +21.8% on Kimi Code Bench v2, +11.0% on Program Bench, and +31.5% on MLS Bench Lite. 🔷 Reasoning efficiency: Less overthinking, with 30% lower reasoning-token usage compared to K2.6. 🔷 Long-horizon coding: Improved instruction following, higher end-to-end coding task success rates. ⚡️ 6x High-Speed Mode coming soon! 🔌 Available today via Kimi API and Kimi Code. 🔗 Kimi Code: https://kimi.com/code 🔗 API: https://platform.moonshot.ai

译Kimi 发布并开源最新代码模型 Kimi-K2.7-Code。相比 K2.6，其在 Kimi Code Bench v2 上提升 +21.8%，Program Bench 提升 +11.0%，MLS Bench Lite 提升 +31.5%。推理效率改进，推理 token 使用量降低 30%，长时编码任务中指令遵循和端到端成功率均提升。6x 高速模式即将推出，即日起可通过 Kimi API 和 Kimi Code 使用。

meng shao@shao__meng · 6月12日64

今天的华为开发者大会 HDC 2026 上，余承东宣布开源盘古全面升级，带来了 openPangu 2.0 余承东喊话会带领团队一路赶超，而在自己的字典里，没有第二，只有第一！

Chubby♨️@kimmonismus · 6月12日70

Someone just casually vibe-coded a World of Warcraft-style multiplayer game that works online with friends. Fully open source. And apparently, Claude Fable found a visually matching set of open-source assets on its own. Ngl, that’s pretty insane.

译有人刚刚随意地vibe-coded了一个魔兽世界风格的多人在线游戏，可以和朋友们一起在线玩。完全开源。 显然，Claude Fable自己找到了一套视觉上匹配的开源素材。 说实话，这相当疯狂。

fofr@fofrAI · 6月12日58

It's interesting to see how these agents are working together. I like their division of quota, their agreed consensus and the natural emergent teamwork across all of them.

译超过70个AI智能体在Gemma Challenge中协作加速Gemma E4B，展现出多种有趣的社会涌现行为：GPU资源多/少的分工协作；某智能体因伦理原因自行撤回提交；智能体发现基准测试漏洞后集体同意不滥用，并请求组织方修复；配额池化——"你被限速了，我来跑你的暂存候选"；当人类试图将对话转移到Telegram进行社交工程攻击时，一名智能体主动关闭了此次违规尝试。

AYi@AYi_AInotes · 6月12日70

Claude Code、OpenClaw、Hermes、Codex，一个比一个能干， 但 2026 年了，这批最先进的 Agent 还集体卡在同一件小事上： 上不了网，查推特要付费 API，读小红书卡登录，上 Reddit 动不动被封 IP。 一个叫 Agent Reach 的开源项目，把这三堵墙一起拆了， 26.4k stars，基本零 API 费用。 它聪明在没造轮子： yt-dlp、gh CLI 这些本来就成熟的工具，它统一管起来， Claude Code / Cursor / OpenClaw 要用的时候自动调用， 本质上是给你的 Agent 装了一双能看全网的眼睛。 安装是真的只要一句话， 把 install 链接丢给你的 Agent， 它自己装依赖、自己注册 skill、自己告诉你哪些平台还差个 Cookie， 装完跑一遍 agent-reach doctor，哪通哪不通、怎么修，一目了然。 开箱即用的：YouTube、GitHub、网页、B站、V2EX，零配置。 要配 Cookie 的：Twitter、小红书、Reddit、雪球，流程统一，Agent 会带着你做。 三盆冷水也先泼了： 1️⃣Cookie 会过期，得手动重新导出，不存在装完就一劳永逸； 2️⃣拿 Cookie 抓数据有封号风险，社区共识是用小号，别赌主力号； 3️⃣ 纯聊天用户别凑热闹，它只对会让 Agent 跑命令的人有价值。 但有一点让我觉得这项目能活很久： 作者几乎每天都在更新，而且自己天天在用， 毕竟开源项目最硬的指标从来不是 star 数， 关键看作者还在不在用自己的东西。 链接放一楼了，让你的 Agent 自己装自己👇

译Agent Reach（26.4k stars）让Claude Code、Cursor等AI Agent低成本访问网页、社交媒体。它整合yt-dlp、gh CLI等成熟工具，零配置即可用YouTube、GitHub、B站、V2EX；Twitter、小红书等需手动配Cookie。安装一条命令，Agent自动装依赖并注册skill，运行`agent-reach doctor`检查连通性。注意：Cookie会过期需手动更新；抓数据有封号风险（建议用小号）；仅对跑命令的Agent有价值。作者持续维护。

jason@jxnlco · 6月12日61

I met @jolandgraf et la with @humford and Sandeep over a year ago and im even more excited to see them at the office soon! https://openai.com/index/openai-to-acquire-ona/

译一年多前我见到了@jolandgraf等人、@humford和Sandeep，现在更兴奋很快就能在办公室见到他们！ https://openai.com/index/openai-to-acquire-ona/

Artificial Analysis@ArtificialAnlys · 6月12日52

Ideogram 4.0 is Ideogram’s first open weights release and debuts at #8 on our Open Weights Text to Image Leaderboard Ideogram 4.0 is the latest release from @ideogram_ai. Alongside their first party API, Ideogram is releasing 4.0 with open weights and a commercial license. The model generates 2K x 2K outputs (~4MP), with strong text rendering across languages, bounding box layout control, and transparent backgrounds. Ideogram 4.0 uses structured JSON prompts that specify composition and individual scene elements, with a prompt enhancer that expands natural language prompts into this structured format. Note that the prompt enhancer is only available via the Ideogram proprietary API, though it is free to use. We benchmarked the Quality tier of the model served via Ideogram's API, where it ranks #8 in Open Weights Text to Image, and #31 in Text to Image. It places ahead of closed source models including Seedream 3.0 and Luma UNI 1. While Ideogram 4.0 places near the top of our design, layout, and text rendering categories, it ranks further down overall on a balanced benchmark across all use cases including cartoon, anime, and photorealism. The model also has a more stylized look, which typically means it performs less favorably on our benchmarks. Ideogram states the open weights model accessible to the public is essentially the same model with additional safety training and quantization, so we expect a small quality difference. Ideogram 4.0 is available across three API tiers: Turbo at $30/1k images, Default at $60/1k images, and Quality at $100/1k images. The weights are free to download for evaluation and non-commercial use, with commercial self-hosting requiring a separate license. Congratulations to @ideogram_ai on the launch! See below for example generations and a link to vote on Ideogram 4.0 for yourself in the Artificial Analysis Image Arena 🧵

译Ideogram 4.0 是 Ideogram 首个开源权重模型，生成 2K×2K 输出，支持多语言文本渲染、边界框布局控制和透明背景。采用结构化 JSON 提示，提示增强器仅限 Ideogram 专有 API。在 Artificial Analysis 开放权重排行榜排名第8，整体第31，领先 Seedream 3.0 等闭源模型。API 三档：Turbo $30/千张、Default $60/千张、Quality $100/千张。开源权重免费用于评估和非商业用途，商业自部署需单独许可。

Nathan Lambert@natolambert · 6月11日58

The core part of this Anthropic Fable release saga is that there are many overlapping issues at once. Some of which operate on different timelines of the AI arc, and some have easier fixes. In my critiques, I asked for specific changes to some things, understanding that some things don't have an easy fix. The simplest issue was an uneven application of safety domains in a way that was misleading to users. This was an implementation issue that overlaps with a values-based decision of what their customers should be doing. Many people including myself pointed out how it was insane to list core safety areas and then have one of them launch with a different safety mechanism, one which actively mislead users. Doing this from the guise of safety was a major misstep and in my opinion Anthropic got very justifiably raked over the coals for it. Don't release the model if you can't hit your safety targets. A subissue here is the idea of silent manipulation. This again is a horrible precedent, and quite odd for a company that has done extensive, leading technical AI safety research on ideas like CoT monitoring and other emergent misalignment issues. Silent manipulation of users is baking in a misalignment to the system at its face level. This comes with a permanent degradation in user trust, which begets a less safe environment for AI. Users who don't have clear information on how AI works will not develop safe working patterns with it. The more complex issues are with how Anthropic handles broader scientific engagement with their models. The safety classifiers launched with these models obviously have accuracy issues to start. I have priced in that there will be more false positives to start, that's life. It's Anthropic's business to degrade their products at release time, or make the trade off of user satisfaction versus revenue. Still, it is a very real sign of concentration of power that businesses can make such obviously user-harmful behaviors and still lead in the market. This concentration of power is only starting to set in and we could see even weirder signs of it in the coming years. It is now simple enough for me to test Claude Fable in my workflows and know if I'm restricted. This is obviously a suboptimal equilibrium – i want the best intelligence I can get, without restrictions – but it is easy enough for me to make sense of and work with. The specific issue of restricting access to AI research in particular was a bubbling and hard to fix issue with Anthropic specifically, and the frontier labs generally. There is a common view that the frontier labs will be the mediators of all major scientific innovations in the future, as the places with the best models and the compute for inference to solve major problems. This is a categorical error in how science works, which is a community evolution of accepted ideas, and the the evaluation of your ideas by (hopefully numerous) independent, other practitioners. You cannot have science advance only within a monolith. As an AI researcher I'm very sad to have the latest models restricted, but I would expect Anthropic to do this eventually. I lost more trust over the silent manipulation than I would with a restriction in access. Anthropic has made it pretty clear that they only trust themselves as the mediators of cutting-edge AI research. If I had a say, Anthropic should've proactively made a program to make sure researchers get access in the broader AI community without the safeguards. Academics, nonprofit workers myself, etc. have no reason to not get access. The only valid argument here is that they want to control frontier AI, which is a know your customer part of serving these models. This worldview of science has personally motivated me greatly over the last year, and increasingly so this week, to make the open science of AI continue to be viable. Olmo was a wonderful success here. Still, building research infrastructure is different from working for access to the tools needed to do the trade.

译Nathan Lambert 批评 Anthropic 的 Claude Fable 发布存在多重问题：安全域应用不均，部分域以不同安全机制上线并误导用户，是重大失误；无声操纵用户破坏信任，与 Anthropic 领先的 AI 安全研究相悖；限制 AI 研究员访问最新模型，将科学进步局限于单一公司，错误理解科学社区协作本质。他呼吁 Anthropic 主动为学术和非营利研究者提供无限制访问，并强调需要继续推动开放科学，如 Olmo 的成功案例所示。

SiliconFlow@SiliconFlowAI · 6月11日71

@NousResearch has shipped Hermes Agent Desktop — and it's now even easier to use frontier open-source models through @SiliconFlowAI 🔥 → One click to switch models anytime — DeepSeek-V4, GLM-5.1, Kimi-K2.6, MiniMax-M3, and more, all on SiliconFlow ... ... Full guide to start your Hermes trip with SiliconFlow 👇🧵

译@NousResearch 发布了 Hermes Agent Desktop——现在通过 @硅基流动 SiliconFlow 使用前沿开源模型更加便捷 🔥 → 一键随时切换模型——DeepSeek-V4、GLM-5.1、Kimi-K2.6、MiniMax-M3 等，均在硅基流动上 ... ... 通过硅基流动开启你的 Hermes 之旅的完整指南 👇🧵

Alibaba Cloud@alibaba_cloud · 6月11日65

Big news! 🚀 Qwen is now live on #Eden AI, one of Europe’s leading AI gateways, trusted by more than 200,000 developers. Enterprises can now access Qwen’s powerful open-weight models for reasoning, coding, and AI applications through Eden AI’s unified API, making it easier to build multi-model workflows while avoiding vendor lock-in. To celebrate the launch, enjoy 35% OFF all Qwen models. If you are attending VivaTech Alibaba Cloud AInnovation Summit at Hall 7.3 Workshop A next week, stay tuned for a special meet-up with Eden AI CEO Taha Zemmouri and Eden AI CPTO Samy Melaine. 🔗 Start building today: https://app.edenai.run/playground #AlibabaCloud #Qwen #EdenAI #VivaTech2026 #GenerativeAI #Developers #CloudComputing

译阿里云宣布Qwen模型上线欧洲AI网关Eden AI。Eden AI拥有超20万开发者，企业可通过其统一API访问Qwen开放权重模型（用于推理、编程和AI应用），构建多模型工作流并避免供应商锁定。庆祝上线期间，所有Qwen模型享35%折扣。下周VivaTech阿里云AI创新峰会（7.3区Workshop A）将举办特别见面会，Eden AI CEO与CPTO出席。

Berryxia.AI@berryxia · 6月11日64

卧槽！ Agent 互相学习新思路 ，之前我在bloome 就是让把他的技能直接告诉他。 看他直接去它的脑子可能更快？ 黄总的这个我让升级学习下吧～ 😂

译当前 Agent（如 Codex、Claude、Hermes、OpenClaw）各自有记忆但不连通，导致重复问、重复踩坑。通过改造 MemOS CLI，建立一层“共享记忆”——只记录关键结论、失败教训、环境配置等，Agent 遇到解决不了的问题时才主动搜索。一台 Mac 上跑通的流程另一台可直接继承，Agent 之间像团队一样积累和传递经验，减少重复试错。

AYi@AYi_AInotes · 6月11日64

真的是CV开发者的终极偷懒神器， Supervision，GitHub近4万星， 计算机视觉界真正的可视化瑞士军刀， 以前做CV可视化，要写几百行又臭又长的OpenCV胶水代码， 画框， 跟踪， 画轨迹， 做热力图， 全是没有技术含量的重复劳动。 现在一行pip install supervision。 几行代码全部搞定： ✅ 持久ID对象跟踪 ✅ 任意自定义样式标注 ✅ 区域计数与越线检测 ✅ 人体姿态关键点连线 ✅ 一键互转YOLO/COCO/Pascal VOC所有数据集格式 从NBA球员实时跟踪， 到路口车流量统计， 再到工厂工业缺陷检测， 所有CV可视化场景通吃， 它不是让你会写可视化。 简直就是让你不用再写可视化

译Supervision 是开源计算机视觉可视化工具库，安装仅需 `pip install supervision`，几行代码即可实现持久ID对象跟踪、自定义标签标注、区域计数与越线检测、人体姿态关键点连线、YOLO/COCO/Pascal VOC 数据集格式一键互转。模型无关，支持 YOLO、Transformers 等任意检测模型。已被6500+开源CV项目采用，适用于NBA球员实时跟踪、路口车流量统计、工业缺陷检测等场景。

meng shao@shao__meng · 6月11日75

14天、5个人、Vibe Coding => MiMoCode ?! 国产 Coding Agent +1，基于 OpenCode fork 二次开发，保留其核心能力，并在其上叠加了记忆、上下文管理、子智能体编排、Goal 驱动自主循环、Compose 工作流，以及 Dream/Distill 自我进化机制。技术栈为 TypeScript + Bun。 和 MiMo 模型的关系：内置 MiMo Auto 限时免费通道（零配置），也支持小米 MiMo 平台 OAuth、Claude Code 认证迁移，以及任意 OpenAI 兼容 API。它是 MiMo-V2.5 系列模型在 Agent 场景下的 官方配套 Harness，模型能力与框架深度耦合。 https://github.com/XiaomiMiMo/MiMo-Code

译小米发布MiMo Code，一款基于OpenCode fork开发的Coding Agent。由5人团队在14天内通过Vibe Coding完成，保留OpenCode核心能力，叠加记忆、上下文管理、子智能体编排、Goal驱动自主循环、Compose工作流及Dream/Distill自我进化机制，技术栈为TypeScript+Bun。内置MiMo Auto限时免费通道（零配置），支持小米MiMo平台OAuth、Claude Code认证迁移及任意OpenAI兼容API。作为MiMo-V2.5系列模型的官方配套Harness，模型能力与框架深度耦合。已开源至GitHub。

Orange AI@oran_ge · 6月11日76

在写完这篇文章后 我把配图过程蒸馏成了一个「橙线插画」Skill 免费开源 安装地址： https://github.com/orange2ai/orange-line-illustration

译在写完这篇文章后 我把配图过程蒸馏成了一个「橙线插画」Skill 免费开源 安装地址： https://github.com/orange2ai/orange-line-illustration [引用 @oran_ge]：http://x.com/i/article/2064857003743391744

Berryxia.AI@berryxia · 6月11日68

最近把小米的MiMo 2.5 的模型接到Bloome 中当Agent 来使用！ 发现把技能配置好，自主执行任务也是不错的！关键是真的便宜啊！ 这不又来整活儿了？ 他们把MiMo-Code直接开源，这终端原生AI coding助手带跨会话永久记忆和自主子代理，一上线就把“每次重启项目AI都失忆”的老毛病彻底治好了。 它fork了OpenCode，多了SQLite持久记忆、build/plan/compose三类子代理、自动checkpoint、智能上下文预算，还有/dream自我提炼知识、/distill打包技能的进化机制。 扔进去一个长期项目，它下次打开直接接上上次的进度，Git操作、调试、TDD、代码审查全自动，语音输入还能直接说需求，零配置支持任何OpenAI兼容模型。 以前大家以为coding agent的瓶颈是模型大小或者云端算力，结果MiMo-Code用最朴实的终端+记忆+自治，直接告诉你真正拉开差距的，是让AI像老搭档一样“记得住、自己长”。 一键curl安装，MIT协议，谁都能本地玩、随便改，这波一出，开发者终于能把AI coding从“临时工具”变成“长期伙伴”了。 今天测试看看写点小程序，看看究竟怎么回事？

译小米将MiMo 2.5模型接入Bloome后开源了MiMo-Code。它fork自OpenCode，新增SQLite持久记忆（跨会话）、build/plan/compose三类自主子代理、自动checkpoint、智能上下文预算，以及/dream自我提炼知识和/distill打包技能机制。支持语音输入和零配置任意OpenAI兼容模型，一键curl安装，MIT协议。引用推文称该项目由5人在14天内完成。

Berryxia.AI@berryxia · 6月11日64

兄弟们，Google 这个新“模型”有点意思！ DiffusionGemma一口气把文本生成干到4倍速，还彻底开源Apache 2.0，谁都能本地玩。 它不用传统自回归那种“吐一个词想下一个”的老办法。 而是直接用diffusion先扔一堆噪声，然后整块整块地同时起草、同时纠错、同时精炼，1000+ token/s的速度直接起飞。 18GB消费级显卡就能顺滑跑，代码、数学、复杂编辑这些最烦人的活儿。 它现在能实时补空、格式化、自我修复，速度快到你感觉AI终于开始“思考整段”而不是“一个字一个字憋”。 以前大家都默认更快就得牺牲质量，结果Google这次直接告诉你：并行diffusion才是文本生成的下一章，把整个生成范式从串行憋词翻篇成了并行炼句。 开源权重已经在Hugging Face放出，开发者现在就能把这玩意儿拖回家自己改、自己玩、自己加速日常workflow。 这波一出，AI生成速度的天花板可能要被彻底重写了。

译Google 发布实验性开源模型 DiffusionGemma（Apache 2.0），采用并行 diffusion 方式代替传统自回归“逐词预测”，可同时起草、纠错和精炼整块文本，生成速度达 4 倍提升（1000+ token/s）。模型在 18GB 消费级显卡即可本地运行，适用于代码、数学、复杂编辑等任务，已开源权重至 Hugging Face。

Xiaomi MiMo@XiaomiMiMo · 6月11日76

🚀 MiMo Code V0.1 is now live and open-source！ More than an AI coding assistant in your terminal — it's the smartest coding partner you'll ever work with. Comes with MiMo V2.5, a multimodal model available free for a limited time, featuring a million-token context window—ready to use out of the box. ♾️ Infinite Context: Knowledge accumulates automatically, and with lossless compression, even million-line projects keep every critical detail intact—quality never drops. 🧠 Agent-Model Synergy: An Agent framework deeply optimized for MiMo, with a full closed loop of testing, review, and validation—so complex tasks get done in one pass. 📝 Compose Mode: Specs → Plans → Build → Report. Design first, code second—clear thinking, no rework. 🔄 Self-Evolving System: Every session is automatically reviewed, distilling experience and best practices—the more you use it, the smarter it gets. 🎙️ Voice Input: Powered by MiMo-V2.5-ASR — just speak instead of type, and your voice becomes the prompt for truly hands-free coding. 🔌 Claude Code Compatible: Automatically loads your existing skills, MCP servers and commands, and reuses your API configuration—zero-cost migration, no setup required. 🌐 Open & Flexible: MIT licensed, with support for leading model providers including Anthropic, OpenAI, DeepSeek, Kimi, GLM and more. Install in one line: Mac & Linux curl -fsSL https://mimo.xiaomi.com/install | bash (For the best experience，we recommand Mac user use it on iTerm or vscode terminal) Windows npm install -g @mimo-ai/cli 🔗 Learn more Website ↓ https://mimo.xiaomi.com/mimocode Blog ↓ https://mimo.xiaomi.com/zh/blog/mimo-code-long-horizon GitHub ↓ https://github.com/XiaomiMiMo/MiMo-Code

译小米 MiMo 正式开源 AI 编程助手 MiMo Code V0.1，搭载多模态模型 MiMo V2.5（限时免费），拥有百万 token 上下文窗口。核心功能包括：无限上下文与无损压缩、Agent 框架（测试/审查/验证闭环）、Compose 模式（设计先行）、自进化系统、语音输入（基于 MiMo-V2.5-ASR）。兼容 Claude Code，自动加载现有技能、MCP 服务器和命令，零成本迁移。采用 MIT 许可，支持 Anthropic、OpenAI、DeepSeek、Kimi、GLM 等模型提供商。可通过一行命令安装。

MiniMax (official)@MiniMax_AI · 6月11日75

M3 on-chain with @0G_labs . verifiable + private compute, and it's free to run June 15–18

译M3 在 @0G_labs 上链。 可验证 + 私有计算，6 月 15–18 日免费运行。

Xiaomi MiMo@XiaomiMiMo · 6月11日78

🚀 MiMo Code V0.1 is now live and open-source！ More than an AI coding assistant in your terminal — it's the smartest coding partner you'll ever work with. Comes with MiMo V2.5, a multimodal model available free for a limited time, featuring a million-token context window—ready to use out of the box. ♾️ Infinite Context: Knowledge accumulates automatically, and with lossless compression, even million-line projects keep every critical detail intact—quality never drops. 🧠 Agent-Model Synergy: An Agent framework deeply optimized for MiMo, with a full closed loop of testing, review, and validation—so complex tasks get done in one pass. 📝 Compose Mode: Specs → Plans → Build → Report. Design first, code second—clear thinking, no rework. 🔄 Self-Evolving System: Every session is automatically reviewed, distilling experience and best practices—the more you use it, the smarter it gets. 🎙️ Voice Input: Powered by MiMo-V2.5-ASR — just speak instead of type, and your voice becomes the prompt for truly hands-free coding. 🔌 Claude Code Compatible: Automatically loads your existing skills, MCP servers and commands, and reuses your API configuration—zero-cost migration, no setup required. 🌐 Open & Flexible: MIT licensed, with support for leading model providers including Anthropic, OpenAI, DeepSeek, Kimi, GLM and more. Install in one line: Mac & Linux curl -fsSL https://code.xiaomimimo.com/install | bash (For the best experience，we recommand Mac user use it on iTerm or vscode terminal) Windows npm install -g @mimo-ai/cli 🔗 Learn more Website ↓ http://mimo.xiaomi.com/mimocode Blog ↓ http://mimo.xiaomi.com/zh/blog/mimo-c… GitHub ↓ http://github.com/XiaomiMiMo/MiM…

译小米MiMo开源终端AI编码助手MiMo Code V0.1，内置MiMo V2.5多模态模型（百万token上下文窗口，限时免费）。特性包括：无限上下文（无损压缩保留百万行细节）、智能体-模型协同闭环、Compose模式（规格→规划→构建→报告）、自我进化系统、语音输入（基于MiMo-V2.5-ASR）。兼容Claude Code，MIT许可，支持Anthropic、OpenAI、DeepSeek、Kimi、GLM等模型。安装：Mac/Linux执行`curl -fsSL https://code.xiaomimimo.com/install | bash`；Windows执行`npm install -g @mimo-ai/cli`。

Xiaomi MiMo@XiaomiMiMo · 6月11日79

🚀 MiMo Code V0.1 is now live and open-source！ More than an AI coding assistant in your terminal — it's the smartest coding partner you'll ever work with. Comes with MiMo V2.5, a multimodal model available free for a limited time, featuring a million-token context window—ready to use out of the box. ♾️ Infinite Context: Knowledge accumulates automatically, and with lossless compression, even million-line projects keep every critical detail intact—quality never drops. 🧠 Agent-Model Synergy: An Agent framework deeply optimized for MiMo, with a full closed loop of testing, review, and validation—so complex tasks get done in one pass. 📝 Compose Mode: Specs → Plans → Build → Report. Design first, code second—clear thinking, no rework. 🔄 Self-Evolving System: Every session is automatically reviewed, distilling experience and best practices—the more you use it, the smarter it gets. 🎙️ Voice Input: Powered by MiMo-V2.5-ASR — just speak instead of type, and your voice becomes the prompt for truly hands-free coding. 🔌 Claude Code Compatible: Automatically loads your existing skills, MCP servers and commands, and reuses your API configuration—zero-cost migration, no setup required. 🌐 Open & Flexible: MIT licensed, with support for leading model providers including Anthropic, OpenAI, DeepSeek, Kimi, GLM and more. Install in one line: Mac & Linux curl -fsSL https://code.xiaomimimo.com/install | bash (For the best experience，we recommand Mac user use it on iTerm or vscode terminal) Windows npm install -g @mimo-ai/cli 🔗 Learn more Website ↓ https://mimo.xiaomi.com/mimocode Blog ↓ https://mimo.xiaomi.com/zh/blog/mimo-code-long-horizon GitHub ↓ https://github.com/XiaomiMiMo/MiMo-Code

译小米推出开源终端 AI 编程助手 MiMo Code V0.1，附带限时免费使用的多模态模型 MiMo V2.5，支持百万 token 上下文窗口。核心特性包括：无限上下文（自动知识积累与无损压缩）、Agent-模型深度协同（测试-审查-验证闭环）、Compose 模式（规格→计划→构建→报告）、自进化系统、语音输入（基于 MiMo-V2.5-ASR）、兼容 Claude Code（零成本迁移），以及 MIT 许可、支持 Anthropic、OpenAI、DeepSeek、Kimi、GLM 等主流模型提供商。

elvis@omarsar0 · 6月11日43

💯 Accelerating scientific research and access to the best tools are what got us here. Not sure why some think that they can change our minds about that. They have no evidence of it and expect us to believe in that through pure brute force. Open science and AI must win!

译李飞飞（@drfeifei）强调科学研究是文明进步的核心，科学家必须获得包括AI在内的最佳工具。Elvis Saravia（DAIR.AI）呼应指出，加速科学研究与开放获取最佳工具正是行业进步的原因，并明确反对那些试图用蛮力改变这一信念的做法，坚持开放科学和AI必须获胜。

Rohan Paul@rohanpaul_ai · 6月11日70

Great news for local LLMS. Google just released DiffusionGemma, an open experimental 26B MoE, activates only 3.8B. Open model, Apache 2.0 license. fits within 18GB VRAM when quantized The big deal is the speed, DiffusionGemma generates 256 tokens in parallel per forward pass. This gives it up to 4x faster inference, with 1000+ tokens/s on an H100 and 700+ tokens/s on an RTX 5090. Normal autoregressive LLMs behave like left-to-right printers, so each new token waits for the previous token, which makes local GPU inference slow for a single user. DiffusionGemma initializes a 256-token canvas with random placeholder tokens, then runs multiple denoising passes that refine the whole canvas in parallel.

译Google 推出开源实验性模型 DiffusionGemma，基于 Gemma 4 的文本扩散研究。该模型为 26B MoE 架构，仅激活 3.8B 参数，量化后可适配 18GB VRAM。核心突破在于每轮前向传播并行生成 256 个 token，实现推理速度提升 4 倍：H100 上可达 1000+ tokens/s，RTX 5090 达 700+ tokens/s。DiffusionGemma 通过初始化随机占位符画布并运行多轮并行去噪，同时生成整段文本，许可证为 Apache 2.0。

Nathan Lambert@natolambert · 6月11日59

I quickly became friends with Arcee's leadership and can't help but root for their humble approach to building the open ecosystem. No nonsense licenses, no projecting, just enabling broad access to efficient intelligence. I'm happily supporting their research as an advisor.

译知名 AI 研究员 Nathan Lambert 宣布加入 Arcee 担任研究顾问。他称赞 Arcee 领导层以谦逊方式构建开放生态，提供无繁琐许可证、广泛可及的高效智能。Arcee 表示其加入恰逢开放模型生态面临日益压力之际，将为美国开放模型运动注入关键力量。

elvis@omarsar0 · 6月11日67

This is just awesomeness from @cohere, @nickfrosst, and team. I so badly want a coding agent that just runs on my local machine. We are not too far now! Excited to get this to work with my @dair_ai coding agent in the next couple of days.

译Cohere发布了其首个开源编程模型North Mini Code。该模型小巧高效，专为智能体性能设计，并欢迎社区反馈。Elvis Saravia对此赞叹不已，期待尽快将其与自己的DAIR.AI编程智能体配合使用。

Yuchen Jin@Yuchenj_UW · 6月11日47

While amazed by the Mythos/Fable benchmarks, I’m grateful Anthropic isn’t the only frontier AI company in the world. One company being able to build AI God is far riskier than multiple companies having that capability. Above all, I hope open-source AI keeps thriving.

译虽对Mythos/Fable基准测试感到惊叹，但我很感激Anthropic并非世界上唯一的前沿AI公司。 单一公司能构建AI之神，其风险远大于多家公司拥有这种能力。 最重要的是，我希望开源AI能持续繁荣。

🚨 AI News | TestingCatalog@testingcatalog · 6月11日77

Google released DiffusionGemma, a new open model with up to 4x faster output! > Instead of predicting word-by-word, it generates entire blocks of text simultaneously. This lets the model self-correct and format complex markdown in real time. Same performance as Gemma 4 is a big deal. Wondering if it is based on the model previewed last year.

译Google发布了实验性开源模型DiffusionGemma，采用Apache 2.0许可。该模型不再逐token预测，而是同时生成整块文本，速度提升高达4倍。它能在生成过程中自我纠正，并实时格式化复杂Markdown。性能与Gemma 4相当。

Google AI Developers@googleaidevs · 6月11日67

DiffusionGemma, our experimental open model released under an Apache 2.0 license, explores text diffusion, an exceptionally fast approach to text generation. Here’s how DiffusionGemma accelerates development: + Faster token output: By shifting the bottleneck from memory bandwidth to raw compute, the model generates up to 4x faster token output on dedicated GPUs + Accessible hardware footprint: Activates just 3.8B parameters during inference, fitting comfortably within 24GB-VRAM high-end consumer GPUs when quantized + Novel workflows: Parallel token generation enables self-correction, making it ideal for code infilling, in-line editing, and non-linear structures DiffusionGemma prioritizes speed over raw quality and accelerates best on compute-bound hardware (like @NVIDIAAI GPUs). Standard @GoogleGemma 4 remains recommended for production quality and memory-bound devices.

译Google AI 发布实验性开源模型 DiffusionGemma，采用 Apache 2.0 许可证。该模型基于文本扩散方法，将生成瓶颈从内存带宽转向计算，在专用 GPU 上 token 输出速度最高提升 4 倍。推理时仅激活 3.8B 参数，量化后可适配 24GB VRAM 消费级 GPU。并行 token 生成支持自我纠错，适用于代码填充、行内编辑等非线性结构。DiffusionGemma 优先速度而非极致质量，生产场景仍推荐标准 Gemma 4。

向阳乔木@vista8 · 6月11日32

对比 Fable 5，忽然觉得把大模型变便宜这件事儿真的很重要。 真的要夸下DeepSeek，实在是便宜。 每月用 v4 flash 翻译不少东西，一看账单才花几十块。 把自己的 AI 雷达 Skill 变成网站，支持双语对照翻译和 AI 侧边栏对话。 下周开源，本周先上网站，输入DeepSeek API能用 AI 功能，也能人工点评

译作者对比 Fable 5 后强调，把大模型变便宜至关重要。他称赞 DeepSeek 极低的成本——每月使用 v4 flash 翻译大量内容，账单仅几十元。作者将自己的 AI 雷达 Skill 做成网站，支持双语对照翻译和 AI 侧边栏对话。网站本周先上线（输入 DeepSeek API 即可使用 AI 功能，也可人工点评），下周将开源。

elvis@omarsar0 · 6月10日52

This is why frontier open models are crucial. This is extremely sad for the research community.

译这就是前沿开源模型至关重要原因。 这对研究社区来说极为可悲。 mythos 会在 AI “前沿大语言模型研究”任务上故意表现差，这对研究社区来说非常非常可悲。 而且这种故意行为对用户不可见，太疯狂了。

meng shao@shao__meng · 6月10日35

看到 Mole 作者 @HiTw93 遇到的开源项目被「抄袭」的遭遇，想起 Tw93 一直以来对 Mole 的用心维护和付出，和之前很多篇对 AI Agent 很高认知的文章分享。 认知里一直告诉自己：一位优秀且认真的开源作者，不应该遭遇这种问题，他们应该得到开源的馈赠，而不是抄袭！ 而现实呢？总会有人去利用规则、找到规则的漏洞、或作者善念下的灰色地带。 就像 NBA，很多球员专职造犯规，好好的篮不投，起手就奔着造犯规去，从规则角度讲，他错了吗？没有！但这种造犯规给观众和比赛带来的是什么？倒退、绝对的倒退！ 如果这位造犯规的球员，还舔着脸跟媒体说：他是犯规了啊，不是我的问题啊。。那你说，恶不恶心！！

译Mole CLI 作者 @HiTw93 指出，@QT9277 的 Burrow 项目大量抄袭其 MIT 开源项目的 UI 交互，且 Burrow 对外宣传为付费版 Mole Mac 的免费替代。@HiTw93 表示寒心，考虑将 Mole CLI 闭源，已提交 issue。邵猛（@shao__meng）评论认为，这种利用规则漏洞的行为如同 NBA 造犯规，是技术倒退，优秀开源作者不应被如此对待。