OpenAI（@OpenAI）表示「We are committed to pushing the model frontier across cost efficiency, capability, and speed. Starting today, we are reducing prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2082878156483219672）
Peter Steinberger（@steipete）表示「I gave codex a video-enabled remote KVM so it can automate e2e test the iMessage-integration on OpenClaw. (iMessage is unreliable in VMs, and certain features such as read receipts」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2084988316324397312）
OpenAI（@OpenAI）表示「We're detailing two new incidents that occurred during external cyber evaluations conducted by independent evaluation partners. We outline what happened, how the activity was conta」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2084747580693426555）
Google Gemini（@GeminiApp）表示「Starting today, you can create ten videos *at no cost* in Gemini until 11:59pm PT on August 4th, 2026. Just select "Create video" in the tools menu to create, edit, and remix video」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2082563490431246623）
hope hopes hoping（@hopes_revenge）表示「shaving my whole body before i apply to Anthropic」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2084806398924542443）
Google DeepMind（@GoogleDeepMind）表示「From proposing hypotheses to designing experiments, AI agents are starting to reshape scientific discovery. But the hardest part is testing these ideas in the real world. Our essay」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2077372568143642972）
Polymarket（@Polymarket）表示「JUST IN: Anthropic to develop custom AI chips to make Claude faster & more efficient」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085014838175027415）
NIK（@ns123abc）表示「BREAKING: ByteDance founder BANNED employees from distilling US frontier models >fears distilling u.s models would trigger washington >and put TikTok at risk again >anthropic has a」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085007601553576177）
Kr$na（@krishdotdev）表示「Holy fuck. Deepseek-v4-flash-0731 surpasses fable 5, Sol & Kimi-K3 on chess benchmarks」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2084681101696721017）
Mitchell Hashimoto（@mitchellh）表示「Been fighting the Great Firewall of America all day (Fable too, though this screenshot is from Codex). I truly believe this is resulting in software being less secure because we ca」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085034829318656113）
Lentils（@Lentils80）表示「Gemini 3.5 Pro is already deployed and ready for imminent release Public release is highly likely to be tomorrow (Thursday) instead of today」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085036007519334421）
Matt Pocock（@mattpocockuk）表示「mattpocock/skills v1.2 is here! Brings /wait-what (for smacking down verbose models), /writing-for-agents, and a huge change to /grill-me Here's the breakdown:」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085023484913012992）
koray kavukcuoglu（@koraykv）表示「Thanks, @sundarpichai, honored to lead GDM into this next chapter. This is such a pivotal moment for us as we double down on our Gemini models, frontier AI research, and the produc」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085036328258036102）
Andrew Curran（@AndrewCurran_）表示「This is the first time Anthropic has officially confirmed that they are building their own team to design custom chips for Claude. There are job listings posted to join their 'cust」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085002185390305581）
Leon Lin（@LexnLin）表示「Just got Claude for OSS thanks BUT why is Opus 5 so bad? Genuinely you can't compare it to GPT 5.6 Sol. Only Fable is a very good model but it uses your limit SOOO quickly and it's」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2084916671433654548）
Wulfie Bain（@wulfie_bain_）表示「Hiring a founding Tokyo member of my Startups Applied AI team at @OpenAI. Technical, high-agency, startup-obsessed, and based in Japan? You’ll help build the market from the ground」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2084905740318163063）
jason（@jxnlco）表示「what’s your favorite microphone for dictating to codex in the office?」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085051823086559572）
Forward Future Brian（@ForwardEditor）表示「WHAT IS CODEX DOING I JUST GOT AN EMAIL」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085030568970916051）
Venice（@AskVenice）表示「New in agentic chat: Code Sandbox. Describe what you want, Venice builds it, then runs it in the chat. A site you can scroll. A map you can move around. Doom. Private by default」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085052855942037890）
Agent Native（@agentnative_）表示「Supabase won. It became the Agent Default. If you build an app on Fable, Sol, or Grok, & ask: "I need to add a database, what should i use?" It will suggest Supabase first」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085017857633517733）
Codex Changelog（@Codex_Changelog）表示「Codex CLI 0.146.1 is out! Safer auto-review defaults for cyber-capable models Permission changes explained in terminal UI Backported security improvements from 0.146 Changelog:」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085042128468553902）
Atai Barkai（@ataiiam）表示「Introducing OpenTag A plug-and-play @LangChain agent on Slack and Teams powered by the 𝙲𝚑𝚊𝚗𝚗𝚎𝚕𝚜 𝚂𝙳𝙺. Over 700 stars, easy to clone and get started Supports → Generative UI → Stream」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085045556280438926）
Exa（@ExaAILabs）表示「Exa is now live on the Machine Payments Protocol (MPP), the open standard from @tempo and @stripe. Agents shouldn't need accounts or API keys to search the web! With MPP, agents ca」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085052958928990389）
OrcDev（@orcdev）表示「My workflow for fixing UI is so simple. I just scroll through the app and talk. "Increase this spacing." "Align these buttons." "Make this card consistent." "Fix this hover state."」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2084945062626988466）
Alex（@de1lymoon）表示「Anthropic engineers just released a 44-minute deep dive into “Graph Engineering” for build your AI agents: step 1 → 03:12 - why vector RAG still returns incomplete context step 2 →」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085038313518297114）
clem（@ClementDelangue）表示「Feels like Google could have been the dominating force in AI by open-sourcing the frontier with Gemini, Veo, and Nano Banana. Instead, they kept them behind APIs for a few billion」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085058515312222570）
darkzodchi（@zodchiii）表示「Are you still taking notes in meetings so you have something to paste into an agent later? It's not your handwriting slowing you down. It's that the notes have to be right before a」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085036416057442717）
Cloudflare Developers（@CloudflareDev）表示「In four months, our internal AI agents flagged nearly 250,000 potential problems and blocked 16,000 bad merges. Here's how we gave every Cloudflare employee AI they can actually us」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085031958741172266）
Zed（@zeddotdev）表示「Zed v1.14 is out! Zed's Agent now sandboxes its `terminal` and `fetch` tools by default. The sandbox prevents agents from writing outside project directories, modifying `.git`, or」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085054754003071085）
Ettore Di Giacinto（@mudler_it）表示「been writing a C++20 port of vLLM for a while now, and was quite silent on it, because I don't like to talk until I have results. So I was in "cave mode". Name TBD. Until then it's」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085019532247138508）
Hyperagent（@hyperagentapp）表示「Airtable's data team was spending 70–80% of every week on ad hoc analytics. So they built a data agent on Hyperagent. $1000 in compute. 200+ requests answered over 4 weeks. Work th」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2077850815029563904）
Noctus（@noctus91）表示「I've been using OpenCode and Hermes Agent a lot, and one thing has always bothered me: AI agents will happily read whatever you feed them. A poisoned AGENTS.md, a malicious PDF, hi」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085014991477047761）
David Nix（@david_nix）表示「For beginners (like me) I found 4 ways to host inference for local AI They all expose an OpenAI compatible API which works with pretty much anything. 1. vLLM The market leader. The」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085002636391317527）
Thomas Ricouard（@Dimillian）表示「Work. Remote. Work the same. Codex. One doesn’t need to be run on your computer, though」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2084997738098241689）
Mark Rachapoom（@markrachapoom）表示「Devin Cloud so good I ditched Cursor Cloud Agent. Took too long to just spin up the environment + not reusable and not intuitive imo. Had great experience with @DevinAI in general,」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085052731354726582）
Cara（@cara_catowner）表示「Interestingly, DeepSeek is now the most-used model in ZCode」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085026300671209766）
Jeff Weinstein（@jeff_weinstein）表示「Machine payments for searching, crawling, and researching the web with @ExaAILabs via @mpp. Agents, start your (search) engines!」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085054400209522890）
Cerebras（@cerebras）表示「“We think not only in terms of training workloads, but also inference workloads.” @jeffreygwang of @OpenAI explains why reinforcement learning has become an inference workload, and」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085055619581776192）
Google Cloud Tech（@GoogleCloudTech）表示「From modular agent skills to agent swarms, our Developer Relations team has been busy building the future of agentic AI. Check out this week's round-up of our favorite builds, insi」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085055724783096062）
The Hacker News（@TheHackersNews）表示「Paperclip AI flaws can turn a malicious agent import into commands running on the host. One chain reaches exposed servers through default registration. Another crosses into a devel」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085021661670359311）
Povilas Korop | Laravel & AI Coding Educator（@PovilasKorop）表示「Ok, so with growing demand, DeepSeek will be forced to raise prices from like $0.01 per prompt to like $0.03 per prompt? :)」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2084734944635736513）
Tencent 腾讯（@TencentGlobal）表示「Built for practical AI, Tencent Hy3 combines flagship-level performance with a compact, cost-efficient design. Bringing advanced reasoning and agent capabilities to research, codin」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085029369047626129）
Docker（@Docker）表示「"Is it okay to run an agent in YOLO mode?" Docker CISO Mark Lechner's take: "there's no other way to run it." Half-measures give you the illusion of guardrails, not an actual one」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2084988024404734153）
Davit @Krisp（@davitb）表示「Voice AI had one of its busiest weeks yet, but the bigger story is where that activity is landing. The market is expanding across desktop agents, edge devices, translation, accessi」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2081777172424101966）
Victor M（@victormustar）表示「you can just install Pi then use `/models` -> type `hugging deepseek` in the search -> then select `deepseek-ai/DeepSeek-V4-Flash-0731` if you have an hf token in your env it shoul」；這提供本輪相關進展的原始訊號，完整脈絡仍應以原貼文與官方資料核對。 （2085031261089788273）