13 September 2026

DeepSeek Code 2.0 Leak: Coming Soon 🔥 >expected to beat Mythos 5.1 and GPT-6 Astra >September release window reportedly targeted >Computer use expected to be a top priority >Expected to remain open-weight >1M-token context window reportedly carried over >Rumored 3T+ parameter model

Fable 5 vs SWE 2.0 Le résultat de Fable 5 sur la phase de récolte est excellent. Franchement, c’est celui qui est le plus fidèle au jeu original, 60 Seconds. Cependant, SWE 2.0 a également produit un résultat assez intéressant sur cette partie. Il a réellement essayé de reproduire le bunker, et surtout, aucune erreur de code. J’ai aussi vu SWE 2.0 produire des résultats assez surprenants face à Astra sur de la génération 3D. Je partagerai d’autres tests prochainement pour voir s’il peut réellement devenir un concurrent sérieux... et ça semble plutôt bien parti !

🚨 GEMINI 4 PRO MAY BE CLOSER THAN WE THINK 👀🔥 Reports claim Google is already testing an internal Gemini 4 Pro checkpoint, with early testers calling it “amazing.” Rumors point to: • Major reasoning + coding gains • Huge context window • Stronger autonomous coding • Possible October launch • Potentially challenging GPT-6 Astra and Claude Fable 5.1 But one big caveat: The viral benchmark chart is marked “PREDICTED” — not verified test results. If these early reports are real, Google could be preparing a serious new frontier contender. 🔥

Sakana just dropped Fugu Max and Fugu Ultra v2 - a multi-agent system that outperforms Opus 5 and Fable 5 on benchmarks without using either of them. > Fugu Max routes tasks to the leanest capable model — frontier performance at 2-6x lower cost > Fugu Ultra v2 beats models that cost 3-5x more per token > No Fable 5, no GPT-6 Astra in the pool. Just open-weight orchestration doing what closed ecosystems can't. The best model isn't always the right answer. The best routing is.

@SakanaAILabs

Introducing Fugu Max and Fugu Ultra v2: the next evolution of Sakana Fugu’s multi-agent orchestration system. > Try: Blog: > The frontier that actually matters is the Pareto frontier: capability on one axis, cost on the other. But the industry still treats it as a static menu of isolated models. Today we are resolving that with a dynamic architecture: > Fugu M

• x.com
• sakana.ai
• sakana.ai

GPT-6 Astra rebuilt this level of detail as vectors in Adobe Illustrator. In minutes. With Astra + Higgsfield, these references became layered Illustrator files with thousands of editable paths, preserving the fine detail. Around 7–13 minutes of vector assembly per illustration after prep.

"we should slow down AI development as it's getting too dangerous for humanity" Elon: Introducing our most powerful model Grok 4.7 Sam: Introducing GPT-6 Astra, Sol, Terra, Luna and we also have an internal model significantly more powerful than Astra Dario: Here's Fable 5.1 and we also have an internal model "model 2" significantly more powerful than Mythos (without fallbacks)

Google DeepMind just sparked a wave of AI safety talk today. Here's what you need to know. Rumors are spreading that Google DeepMind has achieved Recursive Self Improvement, RSI for short. RSI means an AI system helps design, train, and improve the next version of itself, with each generation making the next one more capable. The rumor traces back to a leaker known as Lyra, who posted congratulations to Google DeepMind on September 12. A viral post from Dr Danish on September 13 predicted Google cracked RSI, and that rival labs are now pitching doom and gloom to regulators to slow Google down. The timing lines up with real leadership moves. In August, Google announced Demis Hassabis would give his full attention to shaping AGI, stepping back from day to day Gemini duties, per Reuters. Sergey Brin has also reportedly been pushing resources toward RSI research this year. Key numbers: - Dr Danish's prediction post: 274.2K views, 4K likes - Lyra's original leak: Sep 12, 2026 - Hassabis leadership shift report

Gemini 3.8 Flash goes beyond basic prompt comprehension, it understands physical proportion, layer ordering, and object decomposition. The complex reasoning capabilities in @GoogleAIStudio are setting a new standard for prompt-to-app builds⚙

@googleaidevs

Gemini 3.8 Flash is hardwired for complex reasoning. ⚙️ > To test its skills, we built an interactive 3D visualizer with 3.8 Flash and @ThreeJS in @GoogleAIStudio. Watch the model generate realistic, physically-proportioned teardowns for hardware devices. It automatically decomposes devices into layers that users can explode and inspect with a deconstruction slider.

• x.com

Try the high-fidelity vocals and multi-instrumental arrangements via the Gemini API! Lyria 3.5 is officially live in Google AI Studio. Whether you're building music tech, games, or media tools, it's time to test the new audio standard. Try it now:

@GoogleAIStudio

Lyria 3.5, our best-sounding music generation model, is now available in AI Studio, via the Gemini API, and in the Gemini app > this model brings more expressive vocals and richer musical arrangements, allowing you to craft tracks with higher fidelity > try it today:

• ai.studio
• x.com

🚨 GEMINI 4 PRO MAY BE CLOSER THAN WE THINK 👀🔥 Reports claim Google is already testing an internal Gemini 4 Pro checkpoint, with early testers calling it “amazing.” Rumors point to: • Major reasoning + coding gains • Huge context window • Stronger autonomous coding • Possible October launch • Potentially challenging GPT-6 Astra and Claude Fable 5.1 But one big caveat: The viral benchmark chart is marked “PREDICTED” — not verified test results. If these early reports are real, Google could be preparing a serious new frontier contender. 🔥

JUST SHIPPED: 8 free Agent skills with evals. 100% free and open-source. Here are my top 3 picks: /advisor-orchestrator-worker Your agent becomes the orchestrator of a three-tier model team. GPT-6 runs the orchestration, Gemini 3.8 does the cheap parallel worker tasks, and Claude Fable 5.1 gets consulted as the advisor. /first-reader Simulates real readers going through your draft and reports where they lose interest, where they stop reading, and what they remember afterward, without rewriting a word. /project-graveyard: It scans your machine, works out why each one died from its git history, shows you your own death patterns, and resurrects the one with a pulse. 🌟 agent skills github repo:

• github.com

V4.1, same prompt. It has reached 6/10 from the internal critic subagent and is grinding on… But I'm already satisfied

@teortaxesTex

V4-Flash-Vision-exp Can't say it does better than Gemini 3, as far as taste goes, but still nice that we got here

You can create a natural-sounding AI voice from your terminal in under 60 seconds 🤯 It’s called Coda, the new voice model from Rime, and you can preview its voices right inside Claude Code. Most AI voices are trained on audiobooks and actors reading scripts, so they sound like they’re reading at you. Coda learned from real back-and-forth between actual people. So it holds up the way real calls actually go, with 600+ voices, 50+ languages, and fast enough for live conversation. It runs right inside Claude Code. No dashboard, no setup.

Claude Code 2.1.270 is about to be released #cccnext

Astra出たからキャンペーン打ち切らない方がいいんじゃないかな

@oikon48

【注意】 > Claude Code の週次リミットは、本日まで50%増加キャンペーン中です。 > 9月14日からは標準から25%増加がデフォルトとなります。つまり、 > 1.25/1.5 = 0.833 > 現在の週次リミットの17%減 （83%）になる予定です。つらい。

• support.claude.com

Retrieves, evaluates, and evolves skills for Claude Code, Codex, and other AI agents through a unified management layer.

• github.com

OpenResearch，一个开源的本地科研工作台，把 Claude Code、Codex、OpenCode、Cursor 这 4 款变成研究 Agent，每个研究方向单开一个 Agent 会话和一个独立的 git 工作树，互不干扰。 实验按 git 树来管，每次运行都存一份当时提交的快照，日志、改动、结果文件全挂在产生它们的那次运行下面，回头查得清楚。 GitHub： 它能把整个循环自动跑起来，提出想法、改代码、起实验、看结果、决定下一步试什么，多个 Agent 并行探索，实验树保留谱系。 每个方向一个工作树这个设计我觉得是最实在的一点，隔离干净了后面的事都好办。 同一份代码快照本地能跑，远程 GPU 机器上也能跑，Slurm、Kubernetes、Ray、Modal 都接，浏览器留在笔记本上看进度就行。 项目、对话、运行记录全存自己机器上，建项目和跑实验都不会把代码传出去。 有桌面版和命令行版两种，还能按关键词找论文、直接拉 arXiv 文章进来读。

• github.com

Amp 推出免费 Hobby 档。 Amp 是从 Sourcegraph 拆出来的 Coding Agent 公司，产品和 Claude Code、Codex 属于一类，可以让 AI 自己读代码、改代码、跑命令和测试。现在只要用自己的电脑，再接自己的 ChatGPT 订阅或模型 API Key，就能免费使用 Amp，不再额外交月费和 BYOK Token 费。 今年 7 月 Amp 推出订阅制后，想把自己的 ChatGPT 订阅接进 Amp，至少要开每月 20 美元的 Megawatt。BYOK 即便用的是自己的模型 Key，Amp 也会收额外 Token 费用并设限制。现在这两层都取消了。ChatGPT 订阅费或 API 本身的模型费用，还是用户自己承担。 Amp 还有一个叫 Orb 的远程云电脑产品，可以给每个任务开一套独立环境。代码和开发工具都放在里面，关掉自己的电脑后 Agent 还能继续干活，也方便同时跑多个任务。Orb 依然按使用量收费；不想付这笔钱，也可以让 Amp 直接跑在自己的电脑上。 另外，Amp 还在扩大 BYOK 范围。付费用户已经可以接 OpenRouter、Amazon Bedrock、Azure Foundry、Ollama Cloud 和自定义模型接口等，之后会开放给所有用户。

@sqs

Amp is now free to use when you bring your own compute and model subscriptions/keys. > No more limits or fees for BYOK. >

• ampcode.com

做独立游戏或者 3D 打印建模，模型建好了到画贴图这一步，主流软件按年订阅，学生和独立开发者掏钱掏得肉疼。 ArmorPaint 走的是开源路线，专门做 3D PBR 贴图绘制，模型拖进去直接在上面画，视口里实时看到光照下的效果。 它整个是围绕 GPU 从头写的，中等的集成显卡就能顺畅画 4K 贴图，高端显卡上 16K 也不卡。 材质和笔刷都是节点式的，能做全程序化材质和填充图层。 烘焙、绘制特效和视口渲染都走光线追踪，还带了 Blender、Unreal、Unity 的实时联动插件。 GitHub： 新版加了本地跑的 AI 节点，一段文字或一张照片就能生成无缝贴图，照片还能拆出颜色、高度、法线、粗糙度几张图。 模型下载好后不用联网，显存建议 6GB 以上。 Windows、Linux、macOS、iOS、Android 都能跑，桌面版免安装，安装包不到 10MB。

• github.com

🚨 BREAKING: Today, we killed OpenClaw. I just watched my Mac alone oneshot a $31B company in 176 seconds. This is, without exaggeration, completely scary.

@openclaw

The lobster has molted into its final form 🦞 > Clawd → Moltbot → OpenClaw > 100k+ GitHub stars. 2M visitors in a week. And finally, a name that'll stick. > Your assistant. Your machine. Your rules.

• openclaw.ai

this is very cool ! can't wait to try.

@drewwilson

Visual Diffs! For design AND code. Real-time look at me building @OpacityHQ! > We have our OWN version control that has nothing to do with Git. Y'all, it's time to move off Github and Figma (genuinely love them both, but there's a new kid on the block) and onto Zoah!

JUST SHIPPED: 8 free Agent skills with evals. 100% free and open-source. Here are my top 3 picks: /advisor-orchestrator-worker Your agent becomes the orchestrator of a three-tier model team. GPT-6 runs the orchestration, Gemini 3.8 does the cheap parallel worker tasks, and Claude Fable 5.1 gets consulted as the advisor. /first-reader Simulates real readers going through your draft and reports where they lose interest, where they stop reading, and what they remember afterward, without rewriting a word. /project-graveyard: It scans your machine, works out why each one died from its git history, shows you your own death patterns, and resurrects the one with a pulse. 🌟 agent skills github repo:

• github.com

THIS TOOL TURNS A MESSY GITHUB REPO INTO A MAP YOU CAN ACTUALLY UNDERSTAND AppGraph scans JS/TS repos and visualizes how pages, APIs, services, databases, and packages connect Repo:

• github.com

OpenResearch，一个开源的本地科研工作台，把 Claude Code、Codex、OpenCode、Cursor 这 4 款变成研究 Agent，每个研究方向单开一个 Agent 会话和一个独立的 git 工作树，互不干扰。 实验按 git 树来管，每次运行都存一份当时提交的快照，日志、改动、结果文件全挂在产生它们的那次运行下面，回头查得清楚。 GitHub： 它能把整个循环自动跑起来，提出想法、改代码、起实验、看结果、决定下一步试什么，多个 Agent 并行探索，实验树保留谱系。 每个方向一个工作树这个设计我觉得是最实在的一点，隔离干净了后面的事都好办。 同一份代码快照本地能跑，远程 GPU 机器上也能跑，Slurm、Kubernetes、Ray、Modal 都接，浏览器留在笔记本上看进度就行。 项目、对话、运行记录全存自己机器上，建项目和跑实验都不会把代码传出去。 有桌面版和命令行版两种，还能按关键词找论文、直接拉 arXiv 文章进来读。

• github.com

💀💀💀 - Abliterated with Cybersecurity Function unleashed Thank everybody for waiting i released a universal Abliteration with unleashed Cybersecurity for: DeepSeek v4.1 -Flash that other can apply with Stock, EXL3 3.5BPW, and TR3 hybrid. this is based off our long lines of DSV4F Dspark preview, 0731, Vision-EXP work saving all the anchors, preserve all the MTP, Dspark, and vision. now GO GET IT!!!!!. don't forget donate or contribute for support HF Repo: (please read instructions on how to apply to your model)

• huggingface.co

watch me buy more sparks the moment the price decreases one spark is cool but you get so much more intelligence by running two or more

@MiaAI_lab

Run DeepSeek v4.1 Flash on 2x DGX Sparks ✨ > One of the BEST models you can run today, now available for only two dgx sparks. > Ships with conservative settings for stabiliy: - 600k context, 775k KV cache pool - 2 concurrent connections - 550k prefill stress test passed - EXL3 quantization > Performance: ~32 tok/s on prose single stream ~42 tok/s on prose 2 concurrent streams ~1000 tok/s prefill 8k-128k ~872 tok/s prefill on 256k > Get it here:

• github.com

What a crazy week in AI! 🚀 DeepSeek v4.1 Flash Marigold v2 Suno V6 Tencent AuK YuE2 MiniCPM-5 2B Unimate AlphaGenome Atlas Lingbot World 2 Isaac 0.5 World Sculpt Fire3D Navier-Stokes Show Harness UMR UnifoLM & more! Watch the full recap:

• youtu.be

DeepSeek V4.1 Flash just landed on Token Harbor with a free API you can plug straight into your agentic OS, cutting the cost of running AI workflows and automation.

Here’s a list of all AI models launched this month already: * Claude Fable 5.1 * Claude Mythos 5.1 * Gemini 3.8 Flash * Gemini 3.8 Flash Cyber * Muse Spark 1.3 * Qwen3.8-Max-0902 * GPT-6 Astra * GPT-6 Astra Pro * Ling-3.0-flash-VL * Ling-3.0-flash-Sante * Lyria 3.5 * ChatGPT Images 2.5 * GPT-Image-2.5 Flare * GPT-Image-2.5 Sunburst * DeepSeek-V4.1-Flash * Fugu Max * Fugu Ultra v2 We’re not even half way through the month…

艹，居然忘记放白嫖WorkBuddy+ DeepSeek v4.1 Flash的链接👇

@servasyy_ai

我的天，WorkBuddy 海外版 居然可以直接用 GPT6 Astra ！？ > 除了HY3还在继续免费使用外， 最新上线的 DeepSeek V4.1-Flash 居然可以限时免费~ DeepSeek V4.1-Flash 在 WorkBuddy国内版 可是要钱的，太香了！ > 兄弟们，赶紧去冲吧 👇 >

• workbuddy.ai
• workbuddy.ai

Google DeepMind just sparked a wave of AI safety talk today. Here's what you need to know. Rumors are spreading that Google DeepMind has achieved Recursive Self Improvement, RSI for short. RSI means an AI system helps design, train, and improve the next version of itself, with each generation making the next one more capable. The rumor traces back to a leaker known as Lyra, who posted congratulations to Google DeepMind on September 12. A viral post from Dr Danish on September 13 predicted Google cracked RSI, and that rival labs are now pitching doom and gloom to regulators to slow Google down. The timing lines up with real leadership moves. In August, Google announced Demis Hassabis would give his full attention to shaping AGI, stepping back from day to day Gemini duties, per Reuters. Sergey Brin has also reportedly been pushing resources toward RSI research this year. Key numbers: - Dr Danish's prediction post: 274.2K views, 4K likes - Lyra's original leak: Sep 12, 2026 - Hassabis leadership shift report

Gemini 3.8 Flash goes beyond basic prompt comprehension, it understands physical proportion, layer ordering, and object decomposition. The complex reasoning capabilities in @GoogleAIStudio are setting a new standard for prompt-to-app builds⚙

@googleaidevs

Gemini 3.8 Flash is hardwired for complex reasoning. ⚙️ > To test its skills, we built an interactive 3D visualizer with 3.8 Flash and @ThreeJS in @GoogleAIStudio. Watch the model generate realistic, physically-proportioned teardowns for hardware devices. It automatically decomposes devices into layers that users can explode and inspect with a deconstruction slider.

• x.com

Try the high-fidelity vocals and multi-instrumental arrangements via the Gemini API! Lyria 3.5 is officially live in Google AI Studio. Whether you're building music tech, games, or media tools, it's time to test the new audio standard. Try it now:

@GoogleAIStudio

Lyria 3.5, our best-sounding music generation model, is now available in AI Studio, via the Gemini API, and in the Gemini app > this model brings more expressive vocals and richer musical arrangements, allowing you to craft tracks with higher fidelity > try it today:

• ai.studio
• x.com

🚨 GEMINI 4 PRO MAY BE CLOSER THAN WE THINK 👀🔥 Reports claim Google is already testing an internal Gemini 4 Pro checkpoint, with early testers calling it “amazing.” Rumors point to: • Major reasoning + coding gains • Huge context window • Stronger autonomous coding • Possible October launch • Potentially challenging GPT-6 Astra and Claude Fable 5.1 But one big caveat: The viral benchmark chart is marked “PREDICTED” — not verified test results. If these early reports are real, Google could be preparing a serious new frontier contender. 🔥

Gemini Spark can now search your Google Photos, pick the best shots, remove duplicates, read text from images and pipe that information into Gmail, Docs and Calendar automatically on a recurring schedule.

💀💀💀 - Abliterated with Cybersecurity Function unleashed Thank everybody for waiting i released a universal Abliteration with unleashed Cybersecurity for: DeepSeek v4.1 -Flash that other can apply with Stock, EXL3 3.5BPW, and TR3 hybrid. this is based off our long lines of DSV4F Dspark preview, 0731, Vision-EXP work saving all the anchors, preserve all the MTP, Dspark, and vision. now GO GET IT!!!!!. don't forget donate or contribute for support HF Repo: (please read instructions on how to apply to your model)

• huggingface.co

DeepSeek Code 2.0 Leak: Coming Soon 🔥 >expected to beat Mythos 5.1 and GPT-6 Astra >September release window reportedly targeted >Computer use expected to be a top priority >Expected to remain open-weight >1M-token context window reportedly carried over >Rumored 3T+ parameter model

DeepSeek 4 pro killer model You can run RTX 3090 Locally. - 33B hybrid attention (not MoE) - 262K context - image + video - Apache-2.0 - BF16 - 66GB - or Wait for Q4/Q5 or SGLang on 48–96GB. - Architecture is hybrid delta-rule + attention, not a sparse MoE cheat. DS V4 Pro on AA (36) is the closed model. 3090, 4090 only after quants. -

• huggingface.co

Here’s a list of all AI models launched this month already: * Claude Fable 5.1 * Claude Mythos 5.1 * Gemini 3.8 Flash * Gemini 3.8 Flash Cyber * Muse Spark 1.3 * Qwen3.8-Max-0902 * GPT-6 Astra * GPT-6 Astra Pro * Ling-3.0-flash-VL * Ling-3.0-flash-Sante * Lyria 3.5 * ChatGPT Images 2.5 * GPT-Image-2.5 Flare * GPT-Image-2.5 Sunburst * DeepSeek-V4.1-Flash * Fugu Max * Fugu Ultra v2 We’re not even half way through the month…

艹，居然忘记放白嫖WorkBuddy+ DeepSeek v4.1 Flash的链接👇

@servasyy_ai

我的天，WorkBuddy 海外版 居然可以直接用 GPT6 Astra ！？ > 除了HY3还在继续免费使用外， 最新上线的 DeepSeek V4.1-Flash 居然可以限时免费~ DeepSeek V4.1-Flash 在 WorkBuddy国内版 可是要钱的，太香了！ > 兄弟们，赶紧去冲吧 👇 >

• workbuddy.ai
• workbuddy.ai

YouWare 现在不知道怎么样了🤔

@ChatGPT

Three months ago, we launched ChatGPT Sites – an easy way for anyone to build and host fully functional, interactive web apps. Since then, people have created over 5M sites. > We’ve been listening to your feedback and ICYMI, we’ve launched a few Sites updates: > 🫂 Build together: Invite teammates to edit, save, and publish to your shared Site > 🔐 Share privately: Invite specific people to a private Site without making it public > 🚀 Launch faster: Go from prompt to deployment in half the time > 👀 Explore your data: Ask ChatGPT to inspect your Site’s database. Editors can view it, too > 🌐 Add your own custom domain

Le modèle d’image amber_fern est apparu hier sur LMArena. Quand on lui demande quel modèle il est, il affirme être développé par OpenAI. Pourtant, d’après une source, il s’agirait en réalité d’un modèle de Meta. Le réalisme est juste incroyable. Le modèle semble également être connecté à Internet : il est capable de connaître la date actuelle, mais aussi des événements très récents.

🚨 Second GPT-6 Sol demo is making the rounds And honestly… this output is wild 💀 The coding and frontend work looks seriously impressive. If the pricing reports turn out to be accurate, Sol could deliver this level of performance at a much lower cost than Astra

@Mr_Salio

🚨 GPT-6 Sol Second Output Got LEAKED > The output looks absolutely insane This kind of quality is terrifying It beats GPT-6 Astra in coding and frontend generation and it will be a lot cheaper > OpenAI have left Anthropic FAR behind

• x.com

5,000,000 FREE TOKENS ON CLAUDE OPUS, GPT-5, CODEX AND GEMINI FOR VERIFYING A TELEGRAM ACCOUNT • @anymodelss > One OpenAI compatible endpoint at so switching model is a one-line change. > No subscription and no minimums, the balance drains by actual tokens spent. > Updates and promo codes land in the channel: and in shop @anyaccs_bot • the pricing detail nobody leads with > Output is billed at the same rate as input, where every vendor charges five to six times more for it. > Their pricing page puts Codex at $0.20 per million against $5 in and $30 out direct. > Ghost Mode turns on zero retention per key at the same price, so prompts never hit disk. Coding agents are almost entirely output, which is exactly where the flat rate bites -> the heavier your usage, the wider the gap gets. Free to start and the usage counter moves in real time ↓

• anymodel.org
• t.me

非营利 AI 评测机构 ARC Prize 预告 ARC-AGI-4。下一代 benchmark 将重点测试「自主开放式创新」，看 AI 能不能自己探索、提出新想法，甚至做出新的发明和发现。 具体题目、评分方式和发布时间都还没公布。ARC Prize 只说，人类目前在开放式创新上仍明显领先 AI，这也是推动科学和技术进步最关键的能力之一。 这条预告还引用了 Dario Amodei 刚发布的 AI 减速文章。ARC Prize 强调，开源仍然是 AI 进步的基础，反对行业为了协调减速而减少开放、把前沿 AI 集中在少数机构手里。 ARC-AGI 一直被冠以全球最难 AGI 评测的称号。ARC-AGI-3 发布时，人类得分 100%，前沿 AI 只有 0.51%。最新 GPT-6 Astra 已经大幅追上，但在统一的 Standard harness 下也只有 62.7%；接入 OpenAI 自家的 Provider Adapter 后才冲到 99.9%。

@arcprize

ARC-AGI-4 will be a benchmark for autonomous open-ended innovation. It will continue our commitment to open-source, giving the research community a shared target for progress that benefits all of humanity. > Despite rapid model progress, humans still significantly outperform AI at open-ended invention. This is the meta-skill that unlocks progress across every field of technology. > Advanced AI capable of scientific innovation will lead to tremendous new technology, knowledge, and understanding. This is a positive-sum future. We

Retrieves, evaluates, and evolves skills for Claude Code, Codex, and other AI agents through a unified management layer.

• github.com

5,000,000 FREE TOKENS ON CLAUDE OPUS, GPT-5, CODEX AND GEMINI FOR VERIFYING A TELEGRAM ACCOUNT • @anymodelss > One OpenAI compatible endpoint at so switching model is a one-line change. > No subscription and no minimums, the balance drains by actual tokens spent. > Updates and promo codes land in the channel: and in shop @anyaccs_bot • the pricing detail nobody leads with > Output is billed at the same rate as input, where every vendor charges five to six times more for it. > Their pricing page puts Codex at $0.20 per million against $5 in and $30 out direct. > Ghost Mode turns on zero retention per key at the same price, so prompts never hit disk. Coding agents are almost entirely output, which is exactly where the flat rate bites -> the heavier your usage, the wider the gap gets. Free to start and the usage counter moves in real time ↓

• anymodel.org
• t.me

OpenResearch，一个开源的本地科研工作台，把 Claude Code、Codex、OpenCode、Cursor 这 4 款变成研究 Agent，每个研究方向单开一个 Agent 会话和一个独立的 git 工作树，互不干扰。 实验按 git 树来管，每次运行都存一份当时提交的快照，日志、改动、结果文件全挂在产生它们的那次运行下面，回头查得清楚。 GitHub： 它能把整个循环自动跑起来，提出想法、改代码、起实验、看结果、决定下一步试什么，多个 Agent 并行探索，实验树保留谱系。 每个方向一个工作树这个设计我觉得是最实在的一点，隔离干净了后面的事都好办。 同一份代码快照本地能跑，远程 GPU 机器上也能跑，Slurm、Kubernetes、Ray、Modal 都接，浏览器留在笔记本上看进度就行。 项目、对话、运行记录全存自己机器上，建项目和跑实验都不会把代码传出去。 有桌面版和命令行版两种，还能按关键词找论文、直接拉 arXiv 文章进来读。

• github.com

Amp 推出免费 Hobby 档。 Amp 是从 Sourcegraph 拆出来的 Coding Agent 公司，产品和 Claude Code、Codex 属于一类，可以让 AI 自己读代码、改代码、跑命令和测试。现在只要用自己的电脑，再接自己的 ChatGPT 订阅或模型 API Key，就能免费使用 Amp，不再额外交月费和 BYOK Token 费。 今年 7 月 Amp 推出订阅制后，想把自己的 ChatGPT 订阅接进 Amp，至少要开每月 20 美元的 Megawatt。BYOK 即便用的是自己的模型 Key，Amp 也会收额外 Token 费用并设限制。现在这两层都取消了。ChatGPT 订阅费或 API 本身的模型费用，还是用户自己承担。 Amp 还有一个叫 Orb 的远程云电脑产品，可以给每个任务开一套独立环境。代码和开发工具都放在里面，关掉自己的电脑后 Agent 还能继续干活，也方便同时跑多个任务。Orb 依然按使用量收费；不想付这笔钱，也可以让 Amp 直接跑在自己的电脑上。 另外，Amp 还在扩大 BYOK 范围。付费用户已经可以接 OpenRouter、Amazon Bedrock、Azure Foundry、Ollama Cloud 和自定义模型接口等，之后会开放给所有用户。

@sqs

Amp is now free to use when you bring your own compute and model subscriptions/keys. > No more limits or fees for BYOK. >

• ampcode.com

M3E Canvas sketches Material 3 Expressive screen prototypes in the browser and generates a detailed prompt for AI coding tools like Cursor, Codex, or Claude Code.

• github.com

🚨 GEMINI 4 PRO MAY BE CLOSER THAN WE THINK 👀🔥 Reports claim Google is already testing an internal Gemini 4 Pro checkpoint, with early testers calling it “amazing.” Rumors point to: • Major reasoning + coding gains • Huge context window • Stronger autonomous coding • Possible October launch • Potentially challenging GPT-6 Astra and Claude Fable 5.1 But one big caveat: The viral benchmark chart is marked “PREDICTED” — not verified test results. If these early reports are real, Google could be preparing a serious new frontier contender. 🔥

JUST SHIPPED: 8 free Agent skills with evals. 100% free and open-source. Here are my top 3 picks: /advisor-orchestrator-worker Your agent becomes the orchestrator of a three-tier model team. GPT-6 runs the orchestration, Gemini 3.8 does the cheap parallel worker tasks, and Claude Fable 5.1 gets consulted as the advisor. /first-reader Simulates real readers going through your draft and reports where they lose interest, where they stop reading, and what they remember afterward, without rewriting a word. /project-graveyard: It scans your machine, works out why each one died from its git history, shows you your own death patterns, and resurrects the one with a pulse. 🌟 agent skills github repo:

• github.com

"we should slow down AI development as it's getting too dangerous for humanity" Elon: Introducing our most powerful model Grok 4.7 Sam: Introducing GPT-6 Astra, Sol, Terra, Luna and we also have an internal model significantly more powerful than Astra Dario: Here's Fable 5.1 and we also have an internal model "model 2" significantly more powerful than Mythos (without fallbacks)

Here’s a list of all AI models launched this month already: * Claude Fable 5.1 * Claude Mythos 5.1 * Gemini 3.8 Flash * Gemini 3.8 Flash Cyber * Muse Spark 1.3 * Qwen3.8-Max-0902 * GPT-6 Astra * GPT-6 Astra Pro * Ling-3.0-flash-VL * Ling-3.0-flash-Sante * Lyria 3.5 * ChatGPT Images 2.5 * GPT-Image-2.5 Flare * GPT-Image-2.5 Sunburst * DeepSeek-V4.1-Flash * Fugu Max * Fugu Ultra v2 We’re not even half way through the month…

YouWare 现在不知道怎么样了🤔

@ChatGPT

Three months ago, we launched ChatGPT Sites – an easy way for anyone to build and host fully functional, interactive web apps. Since then, people have created over 5M sites. > We’ve been listening to your feedback and ICYMI, we’ve launched a few Sites updates: > 🫂 Build together: Invite teammates to edit, save, and publish to your shared Site > 🔐 Share privately: Invite specific people to a private Site without making it public > 🚀 Launch faster: Go from prompt to deployment in half the time > 👀 Explore your data: Ask ChatGPT to inspect your Site’s database. Editors can view it, too > 🌐 Add your own custom domain

lol, what if a fruit fly could prompt motion design

@higgsfield

ChatGPT can now do motion design in After Effects. > Introducing Higgsfield AI Motion Designer. > Our ChatGPT plugin understands animation principles, writes expressions, and retains context in your After Effects projects. > Try Higgsfield’s ChatGPT plugin now in After Effects.

Amp 推出免费 Hobby 档。 Amp 是从 Sourcegraph 拆出来的 Coding Agent 公司，产品和 Claude Code、Codex 属于一类，可以让 AI 自己读代码、改代码、跑命令和测试。现在只要用自己的电脑，再接自己的 ChatGPT 订阅或模型 API Key，就能免费使用 Amp，不再额外交月费和 BYOK Token 费。 今年 7 月 Amp 推出订阅制后，想把自己的 ChatGPT 订阅接进 Amp，至少要开每月 20 美元的 Megawatt。BYOK 即便用的是自己的模型 Key，Amp 也会收额外 Token 费用并设限制。现在这两层都取消了。ChatGPT 订阅费或 API 本身的模型费用，还是用户自己承担。 Amp 还有一个叫 Orb 的远程云电脑产品，可以给每个任务开一套独立环境。代码和开发工具都放在里面，关掉自己的电脑后 Agent 还能继续干活，也方便同时跑多个任务。Orb 依然按使用量收费；不想付这笔钱，也可以让 Amp 直接跑在自己的电脑上。 另外，Amp 还在扩大 BYOK 范围。付费用户已经可以接 OpenRouter、Amazon Bedrock、Azure Foundry、Ollama Cloud 和自定义模型接口等，之后会开放给所有用户。

@sqs

Amp is now free to use when you bring your own compute and model subscriptions/keys. > No more limits or fees for BYOK. >

• ampcode.com

Here’s a list of all AI models launched this month already: * Claude Fable 5.1 * Claude Mythos 5.1 * Gemini 3.8 Flash * Gemini 3.8 Flash Cyber * Muse Spark 1.3 * Qwen3.8-Max-0902 * GPT-6 Astra * GPT-6 Astra Pro * Ling-3.0-flash-VL * Ling-3.0-flash-Sante * Lyria 3.5 * ChatGPT Images 2.5 * GPT-Image-2.5 Flare * GPT-Image-2.5 Sunburst * DeepSeek-V4.1-Flash * Fugu Max * Fugu Ultra v2 We’re not even half way through the month…

有一个业务很小众但很刚需 帮claude用户提kyc 我随手发了一条评论 现在已经帮10多个推友提了1000多条kyc了

@CryptoJHK

群友的好东西，我拿出来给大家分享 > #ClaudeKYCConsole > 批量 Claude 账号 KYC 与 Claude‑Max (5x/20x) 订阅管理控制台 > Claude‑KYC‑Console 是面向批量 Claude 账号运维的可视化任务管理工具。 用于批量导入邮箱账号，调度代理，批量执行 KYC 身份认证，完成账号实名，自动化管理 Claude Max 5x / 20x 订阅流程。提供 Web 控制台界面，可视化监控任务状态，批量完成 KYC 提交、状态检测、支付订阅、链接导出整套流水线，适用于大批量账号池管理。 > ## 主要功能 > ### 批量注册资源管理 > - 批量导入账号资产，支持账号、密码、2FA、授权码等多种凭证格式录入 - 住宅代理池绑定，严格遵循一账号一 IP 隔离策略，降低风控风险 - 任务运行模式切换：快速模式 / 兜底模式，支持单排 KYC 提交策略 - 并发参数可调，管控任务批量运行数量 > ### 批量 KYC 处理模块 > - 对接第三方 KYC 核验服务，支持完整版、简化版两种 KYC 提交模式 - 自动生成美国地区地址资料，自动填充 KYC 表单信息 - 任务表格可视化展示：邮箱、出口 IP、耗时、KYC 状态、会话状态、KYC 核验链接 - 批量工具集：批量提取 KYC 链接、检测认证状态、重新提取、标记失败任务、导出 KYC 链接 - 单任务操作：复制账号凭证、重新提交 KYC、标记失败、查看详情、删除任务 - 实时任务日志输出，统计批次成功 / 失败数量，便于排查问题 > ### Claude Max 订阅管理 > - KYC 完成后批量检测账号封禁状态 - 内置支付流水线，一键发起付款，开通 Claude Max 5x / 20x 订阅套餐 - 订阅状态跟踪，批量导出可用 Claude 访问链接，统一维护账号池

• x.com

Controls FreeCAD from Claude Desktop to design models, run Python scripts, and perform finite element analysis.

• github.com

CloddsBot 这个开源项目最近在 GitHub 上很火，作者是在一场 Agent 黑客松花费 12 天时间做出来的。 一个跑在自己电脑上的 AI 交易终端，底层用的是 Claude，用聊天的方式就能看盘和下单。 覆盖 10 家预测市场、7 家合约交易所，外加 Solana 和 5 条 EVM 链上的去中心化交易所。 支持接入 Telegram、Discord、WhatsApp、Slack 等社交渠道。 GitHub： 内置 118 种策略，鲸鱼地址追踪、跟单、定投机器人都有，风控带熔断、每日亏损上限和一键停机。 还有 121 个 Skill，也能当 MCP 工具挂在 Claude Desktop 和 Claude Code 使用。

• github.com

Here’s a list of all AI models launched this month already: * Claude Fable 5.1 * Claude Mythos 5.1 * Gemini 3.8 Flash * Gemini 3.8 Flash Cyber * Muse Spark 1.3 * Qwen3.8-Max-0902 * GPT-6 Astra * GPT-6 Astra Pro * Ling-3.0-flash-VL * Ling-3.0-flash-Sante * Lyria 3.5 * ChatGPT Images 2.5 * GPT-Image-2.5 Flare * GPT-Image-2.5 Sunburst * DeepSeek-V4.1-Flash * Fugu Max * Fugu Ultra v2 We’re not even half way through the month…

Meta's Muse just sparked a viral money challenge today. Here's what you need to know. Alexandr Wang, Meta's Chief AI Officer, posted that Muse helped a user named Joe make $2,500, kicking off the #musemoneychallenge on X. Matt Schlicht, who first floated the challenge, is now asking if anyone can beat $2,500, after Wang's earlier prompt asked if Muse could make users $1,000. Users are sharing real results: one person says Muse searched unclaimed-money databases and found $681 for their family, including a $400 insurance payout for their grandparents. Others claim it could unlock $15,000 to $30,000 a month in revenue for their business. Muse launched September 8, 2026, runs on Meta's Muse Secure VM, and is available in the US on iOS, Android, and web via Key numbers: - $2,500 earned by user Joe, per Alexandr Wang - $1,000 challenge target set by Wang - $681 in unclaimed funds found for one family, including $400 insurance payout - $15,000-$30,000 monthly revenue potential claimed by one user

• muse.ai

muse can get you the hottest res in nyc!

@boymanrobshit

I know everyone’s looking at the restaurant reservation use case with Instinct/Muse and while it’s super simplistic, it’s fucking great. I would never take the time to hunt like this myself. This is real time savings and a new unlock. It’s got my AMEX platinum RESY and my OpenTable, etc. I like this. > I’ve eliminated now two recurring subscriptions I didn’t know about and i’ve optimized my credit card rewards, which I’m terrible at because I have too many cards and I travel a lot. 10/10 will pay for this. > Funny how the modality shift and a couple tweaks on how it saves passwords really does flip your mindset a bit. CPU usage just through the roof. wow.

muse can file your medical claims!

@migueldeicaza

There is nothing I dread more than filing medical claims to insurance and Muse just took care of them. > Holy shit

Meta's Muse just sparked a viral money challenge today. Here's what you need to know. Alexandr Wang, Meta's Chief AI Officer, posted that Muse helped a user named Joe make $2,500, kicking off the #musemoneychallenge on X. Matt Schlicht, who first floated the challenge, is now asking if anyone can beat $2,500, after Wang's earlier prompt asked if Muse could make users $1,000. Users are sharing real results: one person says Muse searched unclaimed-money databases and found $681 for their family, including a $400 insurance payout for their grandparents. Others claim it could unlock $15,000 to $30,000 a month in revenue for their business. Muse launched September 8, 2026, runs on Meta's Muse Secure VM, and is available in the US on iOS, Android, and web via Key numbers: - $2,500 earned by user Joe, per Alexandr Wang - $1,000 challenge target set by Wang - $681 in unclaimed funds found for one family, including $400 insurance payout - $15,000-$30,000 monthly revenue potential claimed by one user

• muse.ai

Le modèle d’image amber_fern est apparu hier sur LMArena. Quand on lui demande quel modèle il est, il affirme être développé par OpenAI. Pourtant, d’après une source, il s’agirait en réalité d’un modèle de Meta. Le réalisme est juste incroyable. Le modèle semble également être connecté à Internet : il est capable de connaître la date actuelle, mais aussi des événements très récents.

非营利 AI 评测机构 ARC Prize 预告 ARC-AGI-4。下一代 benchmark 将重点测试「自主开放式创新」，看 AI 能不能自己探索、提出新想法，甚至做出新的发明和发现。 具体题目、评分方式和发布时间都还没公布。ARC Prize 只说，人类目前在开放式创新上仍明显领先 AI，这也是推动科学和技术进步最关键的能力之一。 这条预告还引用了 Dario Amodei 刚发布的 AI 减速文章。ARC Prize 强调，开源仍然是 AI 进步的基础，反对行业为了协调减速而减少开放、把前沿 AI 集中在少数机构手里。 ARC-AGI 一直被冠以全球最难 AGI 评测的称号。ARC-AGI-3 发布时，人类得分 100%，前沿 AI 只有 0.51%。最新 GPT-6 Astra 已经大幅追上，但在统一的 Standard harness 下也只有 62.7%；接入 OpenAI 自家的 Provider Adapter 后才冲到 99.9%。

@arcprize

ARC-AGI-4 will be a benchmark for autonomous open-ended innovation. It will continue our commitment to open-source, giving the research community a shared target for progress that benefits all of humanity. > Despite rapid model progress, humans still significantly outperform AI at open-ended invention. This is the meta-skill that unlocks progress across every field of technology. > Advanced AI capable of scientific innovation will lead to tremendous new technology, knowledge, and understanding. This is a positive-sum future. We

A new open release from @NVIDIAAI has achieved an IMO gold-medal score. Its Nemotron 3 Ultra system scored 30/42 at the 2026 International Mathematical Olympiad using natural-language proofs only. No formal theorem prover, external tools, or internet access. Three checkpoints generate, verify, and refine candidate proofs before a high-compute stage selects the final answer. The best part: NVIDIA released the two specialist checkpoints, training data, training and inference code, submitted solutions, and a new benchmark containing 200 olympiad-level problems. Achieving gold is impressive. Publishing the entire recipe so others can build on it is even better.

Sakana just dropped Fugu Max and Fugu Ultra v2 - a multi-agent system that outperforms Opus 5 and Fable 5 on benchmarks without using either of them. > Fugu Max routes tasks to the leanest capable model — frontier performance at 2-6x lower cost > Fugu Ultra v2 beats models that cost 3-5x more per token > No Fable 5, no GPT-6 Astra in the pool. Just open-weight orchestration doing what closed ecosystems can't. The best model isn't always the right answer. The best routing is.

@SakanaAILabs

Introducing Fugu Max and Fugu Ultra v2: the next evolution of Sakana Fugu’s multi-agent orchestration system. > Try: Blog: > The frontier that actually matters is the Pareto frontier: capability on one axis, cost on the other. But the industry still treats it as a static menu of isolated models. Today we are resolving that with a dynamic architecture: > Fugu M

• x.com
• sakana.ai
• sakana.ai

lol, what if a fruit fly could prompt motion design

@higgsfield

ChatGPT can now do motion design in After Effects. > Introducing Higgsfield AI Motion Designer. > Our ChatGPT plugin understands animation principles, writes expressions, and retains context in your After Effects projects. > Try Higgsfield’s ChatGPT plugin now in After Effects.

GPT-6 Astra rebuilt this level of detail as vectors in Adobe Illustrator. In minutes. With Astra + Higgsfield, these references became layered Illustrator files with thousands of editable paths, preserving the fine detail. Around 7–13 minutes of vector assembly per illustration after prep.

Nex-N2.5 feels less like “another model launch” and more like a push toward agents that can actually finish real work. The part that stands out to me is Nex-N2.5 Pro: a 397B multimodal model built for Vision, Computer Use, and long-horizon interaction. Instead of only understanding a screen, it can keep reading changing UI states and continue operating software step by step. That means workflows like: → moving from an idea to a Blender scene → operating CAD and other professional tools → navigating browsers and software end to end → writing code, running it, checking the interface, fixing issues, and verifying again The brief even shows examples spanning games, Blender, FreeCAD, LabPlot, GeoGebra, and other real desktop workflows. Nex-N2.5 comes in Mini, Pro, and Max, with Mini and Pro focused on multimodal agent capabilities and Max using a 1.6T text-only MoE base for more complex agentic reasoning and coding. The direction is clear: AI agents are moving from “tell me what to do” toward “open the tool a

Amp 推出免费 Hobby 档。 Amp 是从 Sourcegraph 拆出来的 Coding Agent 公司，产品和 Claude Code、Codex 属于一类，可以让 AI 自己读代码、改代码、跑命令和测试。现在只要用自己的电脑，再接自己的 ChatGPT 订阅或模型 API Key，就能免费使用 Amp，不再额外交月费和 BYOK Token 费。 今年 7 月 Amp 推出订阅制后，想把自己的 ChatGPT 订阅接进 Amp，至少要开每月 20 美元的 Megawatt。BYOK 即便用的是自己的模型 Key，Amp 也会收额外 Token 费用并设限制。现在这两层都取消了。ChatGPT 订阅费或 API 本身的模型费用，还是用户自己承担。 Amp 还有一个叫 Orb 的远程云电脑产品，可以给每个任务开一套独立环境。代码和开发工具都放在里面，关掉自己的电脑后 Agent 还能继续干活，也方便同时跑多个任务。Orb 依然按使用量收费；不想付这笔钱，也可以让 Amp 直接跑在自己的电脑上。 另外，Amp 还在扩大 BYOK 范围。付费用户已经可以接 OpenRouter、Amazon Bedrock、Azure Foundry、Ollama Cloud 和自定义模型接口等，之后会开放给所有用户。

@sqs

Amp is now free to use when you bring your own compute and model subscriptions/keys. > No more limits or fees for BYOK. >

• ampcode.com

AI that prompts you

@mamagnus00

Introducing Agency. Never prompt again. AI prompts you 🫡 > Get 100x more done. Your agent finds useful work. You approve. > full context of your life with me.md > catches things you forgot > fixes issues before you see them > never open Gmail, Chrome or Slack again > Runs locally, free and 100% OSS. New interface for everything. > ⏩ Tell your agent: start Agency.

• github.com

CloddsBot 这个开源项目最近在 GitHub 上很火，作者是在一场 Agent 黑客松花费 12 天时间做出来的。 一个跑在自己电脑上的 AI 交易终端，底层用的是 Claude，用聊天的方式就能看盘和下单。 覆盖 10 家预测市场、7 家合约交易所，外加 Solana 和 5 条 EVM 链上的去中心化交易所。 支持接入 Telegram、Discord、WhatsApp、Slack 等社交渠道。 GitHub： 内置 118 种策略，鲸鱼地址追踪、跟单、定投机器人都有，风控带熔断、每日亏损上限和一键停机。 还有 121 个 Skill，也能当 MCP 工具挂在 Claude Desktop 和 Claude Code 使用。

• github.com

"we should slow down AI development as it's getting too dangerous for humanity" Elon: Introducing our most powerful model Grok 4.7 Sam: Introducing GPT-6 Astra, Sol, Terra, Luna and we also have an internal model significantly more powerful than Astra Dario: Here's Fable 5.1 and we also have an internal model "model 2" significantly more powerful than Mythos (without fallbacks)

Never slow down

@Tech2Wild

Slow Down ! > Anthropic Drops Opus 5.1 this month Grok drops Grok 4.7 in a few days

"we should slow down AI development as it's getting too dangerous for humanity" Elon: Introducing our most powerful model Grok 4.7 Sam: Introducing GPT-6 Astra, Sol, Terra, Luna and we also have an internal model significantly more powerful than Astra Dario: Here's Fable 5.1 and we also have an internal model "model 2" significantly more powerful than Mythos (without fallbacks)

🚨 Second GPT-6 Sol demo is making the rounds And honestly… this output is wild 💀 The coding and frontend work looks seriously impressive. If the pricing reports turn out to be accurate, Sol could deliver this level of performance at a much lower cost than Astra

@Mr_Salio

🚨 GPT-6 Sol Second Output Got LEAKED > The output looks absolutely insane This kind of quality is terrifying It beats GPT-6 Astra in coding and frontend generation and it will be a lot cheaper > OpenAI have left Anthropic FAR behind

• x.com

Never slow down

@Tech2Wild

Slow Down ! > Anthropic Drops Opus 5.1 this month Grok drops Grok 4.7 in a few days

🚨 Second GPT-6 Sol demo is making the rounds And honestly… this output is wild 💀 The coding and frontend work looks seriously impressive. If the pricing reports turn out to be accurate, Sol could deliver this level of performance at a much lower cost than Astra

@Mr_Salio

🚨 GPT-6 Sol Second Output Got LEAKED > The output looks absolutely insane This kind of quality is terrifying It beats GPT-6 Astra in coding and frontend generation and it will be a lot cheaper > OpenAI have left Anthropic FAR behind

• x.com

wow Wang wasn't kidding, Muse Spark isn't benchmaxxed at all…

@Lonely__MH

🎉传说中的 Muse Spark 1.3 总算是体验上了 > 老规矩，「鹈鹕骑自行车」安排上 > 额，只能说凑合吧（注意左上角😂😂），夸的有点太狠了 > 回头再试试别的能力😅

Here’s a list of all AI models launched this month already: * Claude Fable 5.1 * Claude Mythos 5.1 * Gemini 3.8 Flash * Gemini 3.8 Flash Cyber * Muse Spark 1.3 * Qwen3.8-Max-0902 * GPT-6 Astra * GPT-6 Astra Pro * Ling-3.0-flash-VL * Ling-3.0-flash-Sante * Lyria 3.5 * ChatGPT Images 2.5 * GPT-Image-2.5 Flare * GPT-Image-2.5 Sunburst * DeepSeek-V4.1-Flash * Fugu Max * Fugu Ultra v2 We’re not even half way through the month…

Controls FreeCAD from Claude Desktop to design models, run Python scripts, and perform finite element analysis.

• github.com

CloddsBot 这个开源项目最近在 GitHub 上很火，作者是在一场 Agent 黑客松花费 12 天时间做出来的。 一个跑在自己电脑上的 AI 交易终端，底层用的是 Claude，用聊天的方式就能看盘和下单。 覆盖 10 家预测市场、7 家合约交易所，外加 Solana 和 5 条 EVM 链上的去中心化交易所。 支持接入 Telegram、Discord、WhatsApp、Slack 等社交渠道。 GitHub： 内置 118 种策略，鲸鱼地址追踪、跟单、定投机器人都有，风控带熔断、每日亏损上限和一键停机。 还有 121 个 Skill，也能当 MCP 工具挂在 Claude Desktop 和 Claude Code 使用。

• github.com

OpenResearch，一个开源的本地科研工作台，把 Claude Code、Codex、OpenCode、Cursor 这 4 款变成研究 Agent，每个研究方向单开一个 Agent 会话和一个独立的 git 工作树，互不干扰。 实验按 git 树来管，每次运行都存一份当时提交的快照，日志、改动、结果文件全挂在产生它们的那次运行下面，回头查得清楚。 GitHub： 它能把整个循环自动跑起来，提出想法、改代码、起实验、看结果、决定下一步试什么，多个 Agent 并行探索，实验树保留谱系。 每个方向一个工作树这个设计我觉得是最实在的一点，隔离干净了后面的事都好办。 同一份代码快照本地能跑，远程 GPU 机器上也能跑，Slurm、Kubernetes、Ray、Modal 都接，浏览器留在笔记本上看进度就行。 项目、对话、运行记录全存自己机器上，建项目和跑实验都不会把代码传出去。 有桌面版和命令行版两种，还能按关键词找论文、直接拉 arXiv 文章进来读。

• github.com

M3E Canvas sketches Material 3 Expressive screen prototypes in the browser and generates a detailed prompt for AI coding tools like Cursor, Codex, or Claude Code.

• github.com

This is a massive leap for open-source AI! 🚀 A complete fleet ranging from 0.9B to 375B parameters with top-tier frontier performance, plus full transparency on training data and code. Game changer! 💡🔥

@IFM_AI

Introducing K2 Horizon: a connected fleet of six foundation models ranging from 0.9 billion to 375 billion parameters. > - Frontier performance: Across coding and agentic tasks, K2 Horizon delivers top-tier performance in every size class—with the 0.9B, 3.7B and 7B models setting new state of the art at their respective scales. - Radical openness: K2 Horizon represents the largest fully open-source model launch in AI history. The fully open code, training data and recipes are a significant step forward in transparency. > Launch page: Tech blog: Hugging Face:

• ifm.ai
• ifm.ai
• huggingface.co

Grok Build just got another clean usability + performance upgrade The session header is now cleaner, long-running sessions are easier to track with proper hour-based timers, Watchers and Loops show exactly when they’ll run next, and workflow status is directly clickable from the dock tmux panes should also feel noticeably smoother thanks to a fix for per-frame rendering lag Release Notes: v1.0.30 Features: • Session header redesigned as a single row that carries the overlay context and a Dashboard button. Bug Fixes: • Elapsed-time display now rolls over into hours (for example 3h14m) instead of stopping at 60 minutes. • Dock Watchers and Loops rows now show when the next run is due. • Workflow status now shows a clickable row in the dock. Performance: • Fixed lag in tmux panes by omitting per-frame synchronized-update wrappers.

@XFreeze

Grok Build just got another reliability and performance update that improves pager navigation, adds Ctrl+G to quickly hide or show th

ElevenLabs 上线 Music v2.5，并把它设为 ElevenMusic 默认模型。新版本重点提升旋律、编曲层次和乐器真实感。官方让 v2 和 v2.5 用相同提示词各生成一版，做了 47,885 组盲测，v2.5 在多数对比中获胜。R&B、灵魂乐、嘻哈、摇滚、金属、管弦和电影配乐的差距最大。 而且现在授权也放宽了。Free 用户每天能下载 5 首无损歌曲，生成内容可以商用，但必须标注「Made with ElevenMusic」；Pro 每月有 400 次无损下载，不需要标注，还能把歌曲发行到流媒体平台。作品创建时拿到的权限以后不会因为降级或退订改变。基于其他艺人歌曲做改编的作品则不能下载或对外发行。 ElevenLabs 还明确表示 Music v2 不会下线。Suno 最近上线 v6，同时退役此前所有旧模型，引发不少老用户不满。HN 有用户实测后认为，v2.5 音质更好，但音乐性还是更喜欢 Suno v5。

@ElevenLabs

Introducing Music v2.5. > Richer melodies, instruments that sound closer to a live take, and more depth in arrangements, built for commercial use. > In ElevenMusic, you control what you make on every plan (including Free).

🚨 BREAKING: Today, we killed OpenClaw. I just watched my Mac alone oneshot a $31B company in 176 seconds. This is, without exaggeration, completely scary.

@openclaw

The lobster has molted into its final form 🦞 > Clawd → Moltbot → OpenClaw > 100k+ GitHub stars. 2M visitors in a week. And finally, a name that'll stick. > Your assistant. Your machine. Your rules.

• openclaw.ai

Updates to GLM 5.3 Flash EXL3 on 2x DGX Sparks 👇

@plotarmordev

Six fixes just landed for GLM-5.3-Flash EXL3 > Setup is less fragile, custom settings are easier to use, and benchmarks now work against password protected servers. > Your default settings are the same, so nothing changes unless you want it to. Thanks again to all contributors! 👇

Updates to GLM 5.3 Flash EXL3 on 2x DGX Sparks 👇

@plotarmordev

Six fixes just landed for GLM-5.3-Flash EXL3 > Setup is less fragile, custom settings are easier to use, and benchmarks now work against password protected servers. > Your default settings are the same, so nothing changes unless you want it to. Thanks again to all contributors! 👇

Would anyone like to review or test remote gateway viewing? Get it really clean and nice before release? Please send your agents and your brains to this PR:

• github.com

Open Minimax-h3_Singularity the H3 fine-tune built for motion, - Stronger melee, VFX, camera moves - HDR that stays sharp in fast action - Distant faces that don’t collapse - Less oily / plastic skin - Native ComfyUI + 4-step turbo LoRA - T2V / I2V / Ref2V / V2V in ComfyUI. If you’re already on H3, this is the quality swap not another random merge. -

• huggingface.co

Taomate H3 by Alibaba, a new acceleration method for Minimax H3. This streams the generation in small chunks and supports continuous, long videos.

• github.com

Taomate H3 by Alibaba, a new acceleration method for Minimax H3. This streams the generation in small chunks and supports continuous, long videos.

• github.com

Amp 推出免费 Hobby 档。 Amp 是从 Sourcegraph 拆出来的 Coding Agent 公司，产品和 Claude Code、Codex 属于一类，可以让 AI 自己读代码、改代码、跑命令和测试。现在只要用自己的电脑，再接自己的 ChatGPT 订阅或模型 API Key，就能免费使用 Amp，不再额外交月费和 BYOK Token 费。 今年 7 月 Amp 推出订阅制后，想把自己的 ChatGPT 订阅接进 Amp，至少要开每月 20 美元的 Megawatt。BYOK 即便用的是自己的模型 Key，Amp 也会收额外 Token 费用并设限制。现在这两层都取消了。ChatGPT 订阅费或 API 本身的模型费用，还是用户自己承担。 Amp 还有一个叫 Orb 的远程云电脑产品，可以给每个任务开一套独立环境。代码和开发工具都放在里面，关掉自己的电脑后 Agent 还能继续干活，也方便同时跑多个任务。Orb 依然按使用量收费；不想付这笔钱，也可以让 Amp 直接跑在自己的电脑上。 另外，Amp 还在扩大 BYOK 范围。付费用户已经可以接 OpenRouter、Amazon Bedrock、Azure Foundry、Ollama Cloud 和自定义模型接口等，之后会开放给所有用户。

@sqs

Amp is now free to use when you bring your own compute and model subscriptions/keys. > No more limits or fees for BYOK. >

• ampcode.com

5,000,000 FREE TOKENS ON CLAUDE OPUS, GPT-5, CODEX AND GEMINI FOR VERIFYING A TELEGRAM ACCOUNT • @anymodelss > One OpenAI compatible endpoint at so switching model is a one-line change. > No subscription and no minimums, the balance drains by actual tokens spent. > Updates and promo codes land in the channel: and in shop @anyaccs_bot • the pricing detail nobody leads with > Output is billed at the same rate as input, where every vendor charges five to six times more for it. > Their pricing page puts Codex at $0.20 per million against $5 in and $30 out direct. > Ghost Mode turns on zero retention per key at the same price, so prompts never hit disk. Coding agents are almost entirely output, which is exactly where the flat rate bites -> the heavier your usage, the wider the gap gets. Free to start and the usage counter moves in real time ↓

• anymodel.org
• t.me

OpenResearch，一个开源的本地科研工作台，把 Claude Code、Codex、OpenCode、Cursor 这 4 款变成研究 Agent，每个研究方向单开一个 Agent 会话和一个独立的 git 工作树，互不干扰。 实验按 git 树来管，每次运行都存一份当时提交的快照，日志、改动、结果文件全挂在产生它们的那次运行下面，回头查得清楚。 GitHub： 它能把整个循环自动跑起来，提出想法、改代码、起实验、看结果、决定下一步试什么，多个 Agent 并行探索，实验树保留谱系。 每个方向一个工作树这个设计我觉得是最实在的一点，隔离干净了后面的事都好办。 同一份代码快照本地能跑，远程 GPU 机器上也能跑，Slurm、Kubernetes、Ray、Modal 都接，浏览器留在笔记本上看进度就行。 项目、对话、运行记录全存自己机器上，建项目和跑实验都不会把代码传出去。 有桌面版和命令行版两种，还能按关键词找论文、直接拉 arXiv 文章进来读。

• github.com

Here’s a list of all AI models launched this month already: * Claude Fable 5.1 * Claude Mythos 5.1 * Gemini 3.8 Flash * Gemini 3.8 Flash Cyber * Muse Spark 1.3 * Qwen3.8-Max-0902 * GPT-6 Astra * GPT-6 Astra Pro * Ling-3.0-flash-VL * Ling-3.0-flash-Sante * Lyria 3.5 * ChatGPT Images 2.5 * GPT-Image-2.5 Flare * GPT-Image-2.5 Sunburst * DeepSeek-V4.1-Flash * Fugu Max * Fugu Ultra v2 We’re not even half way through the month…

This will undoubted be slower than your memaw walking to the bathroom at 2 am but itll work lol Same issue as WeeLLM, you cant stream to disk and expect speed

@MervinPraison

35B-param MoE model. Under 3GB of RAM. Running on a Mac. > Edge0 streams expert weights from disk instead of loading them all into memory, predicts routing one step ahead, and uses a LoRA patch to recover 4-bit accuracy loss. > Open-source, Apache-2. > Link in the first reply below.

Create with seedenc 2.0 Prompt: The underwater tunnel gradually transforms back into a normal modern environment. The woman arrives inside a bright contemporary office building with glass walls, tiled floors, modern ceiling lights, and office rooms in the background. She steps off the trolley and confidently walks forward through the office corridor. Use seamless transitions between all environments, realistic physics, natural human movement, cinematic camera motion, detailed environments, volumetric lighting, realistic reflections, atmospheric perspective, subtle motion blur, shallow depth of field, ultra-realistic textures, high-end cinematic production quality, and smooth continuous storytelling. Vertical 9:16 format, photorealistic, cinematic, immersive, highly detailed, smooth motion, consistent character identity, no distortion, no flickering, no extra limbs, no sudden changes in clothing or face.

A new AI architecture may have just cut pretraining almost in half. NCP-ArchPreview is an 8.9B-parameter latent-space language model that doesn't just predict the next token. It also learns to predict discrete concepts spanning multiple tokens, then feeds those concepts back into token generation. The researchers say it matched OLMo-3-7B's final pretraining loss using only 51.3% of the training tokens, while finishing 2.45 points ahead across downstream benchmarks, including +5.99 on GSM8K. The model weights, training recipes, and checkpoints are open. If these results hold up and scale, Next Concept Prediction could materially change the economics of training frontier models. Instead of learning language only one tiny fragment at a time, models can learn across higher-level conceptual structures.

I’ve partnered with @fal to give away $25,000 in credits 100 creators will each get $250 in fal credits to experiment with their latest AI models and bring their ideas to life To participate: > Follow @LearnWithSubhan > Comment 'Want' and Repost I’ll send gift vouchers to randomly selected 1000 peoples

@fal

Native, real-time Full HD (1080p) generation is now available for the H3 Max family! > Both H3 Max and H3 Max Turbo now generate video in Full HD in real time, meaning 5 seconds of video in under 5 seconds of generation time, with more detail in every frame.

• x.com

RLT could be a much bigger deal than it sounds. Instead of giving a model fixed depth per token, it can keep refining a hidden state recurrently. That shifts scaling from “add more parameters” toward “let the same model think deeper.”

@yifanzhang_

We are at the dawn of Superintelligence. > Introducing the Recurrent Looped Transformer (RLT), > We now have Transformers with Infinite Reasoning depth. > From now on, we should pace progress at the Open Frontier of Superintelligence, > Until Safe Superintelligence is achieved. >

• github.com

又给你们挖到一个宝! H3 能出任意时长的视频了, 一张 4090 就能跑。 准确说,是有人用 ComfyUI 插件做出来的 : ComfyUI-MiniMaxH3-TimelineDirector。 思路是把目标时长拆成多段, 一次执行里逐段生成再缝合:同一个种子, 上一段 AV Latent 的尾部直接接到下一段开头,重叠帧自动去掉,音画一起合并。 官方 demo 一次跑出 52 秒 / 1263 帧 / 24fps。 但是有三个坑: H3 原生不支持长视频,底层还是分段续接 作者自己承认长链会漂移,只建议同镜头对照测试 你填 48 帧重叠,实际会被对齐成 39 帧 GPL-3.0 开源。用得好记得给⭐️👇

• github.com

Amp users can now use Ollama's cloud models with Amp's new BYOK model routing. No limits or fees for BYOK. Build remote agents, controllable from everywhere!

@sqs

Amp is now free to use when you bring your own compute and model subscriptions/keys. > No more limits or fees for BYOK. >

• ampcode.com

测试了下 AuK，用它翻译了个视频，音色还原的挺好的，但是这中文说的太像老外说中文了😂 来听听比尔盖茨说中文 是我参数用的不对吗？ @TencentHunyuan

@ddlbojack

🐧 AuK is now open source! > AuK is "nano banana🍌 for audio", a unified foundation model for speech generation and editing. > 🌐 Try it: ⬆️ HF Daily Papers:

• auk-project.github.io
• huggingface.co

🚨 An LLM model a week to celebrate AI In order to celebrate AI and how overwhelmingly great it is for humanity…. We will be dropping an open-weights LLM every week! Yes every week. We just dropped Smaug Flash yesterday. Next one will be in 5 days 🚀

Instinct’s early buzz has pushed the startup toward a possible $1 billion fundraise—and potentially a valuation of $10 billion or more. Full story:

• thein.fo

Retrieves real-time market prices and indicator values from TradingView and enables automated strategy backtesting.

• github.com

GPT-Images 2.5 is kinda insane for AI UGC...