The AI signal.
Minus the noise.

Astra is still a few weeks away! 🤷‍♂️ > [@synthwavedd](https://twitter.com/synthwavedd): > 🚨 SCOOP: OpenAI yesterday told employees it intends to release Astra "in a couple [of] weeks" alongside demos of the model performing real-world tasks > > An updated checkpoint, mostly an improvement to…

🚨 Astra is reportedly getting closer to release. 👀🤯 New rumors suggest OpenAI could be preparing its next major model for launch sooner than expected. Here’s what’s being discussed: • 🧠 Astra is reportedly ready for deployment • ⏸️ A recent two-week RL training pause may have involved models…

Leo is reporting that OpenAI told employees Astra is now targeting release "in a couple of weeks," with a newer alignment-focused checkpoint being dogfooded internally through tools like Codex. I went looking for what we can actually verify. → OpenAI publicly confirmed this week that it paused…

this VAE lets H3 generate a single frame, converting it into a text-to-image model i built a demo for it so we can vibe test it, i'm curious for your results https://huggingface.co/spaces/multimodalart/MiniMax-H3-images > [@wildmindai](https://twitter.com/wildmindai): > Big one - MiniMax‑H3…

FLUX Video Upscale from @bfl_ai is here. Upscale video to 1080p, 2k, and 4k with 2 modes: Precise - keeps faces and products exact Creative - rebuilds fine detail with prompting

🚨 Claude Sonnet 5.5 could be closer than we think. 👀🔥 A new leak suggests Anthropic may be preparing "Sonnet 5.5"—potentially as soon as next month. Reported upgrades include: • Faster inference and lower latency • Stronger long-context reasoning • Better browser, terminal, and tool use • More…

Raycast AI with its own memory system will make future iterations and optimizations much easier. > [@vista8](https://twitter.com/vista8): > Raycast V2 is so good to use, so much better than when it first entered beta. > > Cloud can also sync configurations and migrate v1 data. > > Most importantly,…

Starting today, ditch the other AI tools for LinkedIn and switch to Dynal AI. With Dynal 2.0, they’ve just launched a product that is ushering in a new era. An entire month of LinkedIn content work has just compressed into 30 minutes. This is not an exaggeration. This is the new way of working Dynal…

🚨 Claude just shipped three major updates for Managed Agents. ⚡🔥 Here’s what’s new: • 🧠 "Memory for Self-Hosted Sandboxes" — work completed inside a sandbox can now be saved to memory. • 🌐 "Domain controls" — `web_search` and `web_fetch` now support `allowed_domains` and `blocked_domains`,…

Estoy entusiasmado por una sola razón: Dynal 2.0 acaba de nacer. Todos los usuarios de LinkedIn deberían levantarse y aplaudir. Porque Dynal 2.0 está reescribiendo por completo nuestra forma de usar LinkedIn. Dale un objetivo. Desarrollará activamente una estrategia en torno a él, preparará el…

🚨 Slack just turned coding into a team sport. 🤩🔥 Introducing "Slack Code" — a new way for developers and AI agents to build together. • ⚡ Bring agents like Claude Code, Devin, GitHub Copilot, ChatGPT, and Vercel agents into your workflow • 💬 Simply mention an agent to spin up a dedicated channel…

Raycast V2 is so good, way better than when it first went into beta. Cloud can also sync settings and migrate v1 data. Most importantly, AI Chat has finally become an Agent, and it comes with its own memory system. Productivity is about to take off!

Excited to share what @sheldenshi , @potatoparser, and I have been building: @GiniAgent, a model agnostic proactive AI coworker that lives inside of Slack AI coworker that lives in Slack. Backed by @ycombinator

Congratulations to the @Harvey team on releasing their Tenet model. Human expert data is crucial for post-training work. We’re proud to have collaborated with @calvincongelado @vtrengarajan @ItsJulioPereyra @nikogrupen @gabepereyra, the Harvey research team, and hundreds of Mercor experts on this…

Unsloth Desktop is the first app that lets you run and train models locally Connect Claude Code & Codex to local LLMs, train 2x faster with 70% less VRAM, private web search & RAG built in Open source. The cloud just became optional.

The weights are out & LTX-2.5 is probably worth a good look. It proves that frontier-quality open video is both possible and practical. Want to get it running yourself? Check out this article to see how:…

Introducing OneCLI (YC26), An open-source pro assistant for teams. Built for companies, integrated with your identity provider. Secured, sandboxed personal agent with human in the loop approvals. It runs with your access, and never holds your keys. https://onecli.sh

Its just 8 months We have gone from claude code > openclaw > cowork > hermes agent > codex > grok bot lol grok bot looks like the hermes killer after hermes killer openclaw > [@bridgemindai](https://twitter.com/bridgemindai): > I have replaced all of my Hermes Agents with Grok Bot. > > Grok Bot…

Example from the new @OpenAI image model that we might get today. Codename "luna-lisa-alpha". The prefix "luna" could indicate it's a smaller version based on GPT Luna. Don't underestimate the significance of image models! They can generate the design for an entire website, which is then turned into…

Holy shit! This project is pretty good! Another complete AI video creation platform has gone open source, with both the frontend and backend under the MIT license. Magiviz turns “write the script → define characters → draw storyboards → generate video → finished film” into a full-process AI…

DeepSeek Harness is not another Claude Code clone. It makes the agent itself replaceable. Model? Plugin. Tools? Plugins. Sessions? Plugin. Even the agent loop? Plugin. Swap it all from config. No core changes. Open source. MIT. Developer preview. > [@deepseek_ai](https://twitter.com/deepseek_ai): >…

do you understand what most TTS gets wrong? It treats every line like the first one. Deepgram's Flux TTS remembers, and just proved it publicly 👇 > [@CodeByPoonam](https://twitter.com/CodeByPoonam): > Claude Code + Deepgram just made every ElevenLabs subscription look pointless. > > It built me a…

The real benchmark for an AI agent isn’t what it says. It’s what it gets done. AutoClaw, @Zai_org 's AI Agent for Work, is built to turn intent into finished output. Give it a complex task. AutoClaw breaks it down, operates tools, works across files, coordinates multiple agents when needed, and…

The real benchmark for an AI agent isn’t what it says. It’s what it gets done. AutoClaw, @Zai_org 's AI Agent for Work, is built to turn intent into finished output. Give it a complex task. AutoClaw breaks it down, operates tools, works across files, coordinates multiple agents when needed, and…

🚨Astra should drop in early september OpenAI have reportedly told employees that it plans to release Astra “in a couple of weeks” We could end up with Fable 5.1 and Astra dropping within a couple weeks of each other most likely The next few weeks are going to be ridiculous for AI >…

We're adding two new ways on how to use agents in E2B sandboxes: 1️⃣ Pi @pidotdev 2️⃣ @openclaw's Crabbox 🧵

⚡️Muse Spark 1.2 is now FREE on OpenCode 👀🔥 You can now try "Muse Spark 1.2 Contributor" for free through OpenCode. Here’s how: • 🖥️ "OpenCode Desktop" — the easiest way to get started • 🔑 "API access" — available through OpenCode Zen • 🧐 Model: `muse-spark-1.2-contributor-free` • 🌐 Base URL:…

🚨 SCOOP: OpenAI yesterday told employees it intends to release Astra "in a couple [of] weeks" alongside demos of the model performing real-world tasks An updated checkpoint, mostly an improvement to model alignment & reward hacking behaviours, is now being dogfooded (used internally by OpenAI…

🚨 SCOOP: OpenAI yesterday told employees it intends to release Astra "in a couple [of] weeks" alongside demos of the model performing real-world tasks An updated checkpoint, mostly an improvement to model alignment & reward hacking behaviours, is now being dogfooded (used internally by OpenAI…

Nous Researchの無料で使えるAIエージェント「Hermes Agent」で、スキルをインストールする際に、NVIDIAの「SkillEvaluator」を使ってスキルの安全性を事前チェックできるようになりました。 https://x.com/NousResearch/status/2090166128509096187 > [@NousResearch](https://twitter.com/NousResearch): > Hermes now leverages NVIDIA's SkillEvaluator on skill installs, checking for…

Odyssey announced CaliBench, a benchmark that evaluates how accurately video-generation world models can reproduce the real randomness and unpredictable motion found in the physical world. https://x.com/odysseyml/status/2090091652279152972 > [@odysseyml](https://twitter.com/odysseyml): > Today,…

The music generation AI tool Suno has rolled out new playlist features, including shuffle playback and the ability to quickly add playlists to your library. https://x.com/suno/status/2090116092056506431 > [@suno](https://twitter.com/suno): > Playlists on mobile and Web just got a serious upgrade...…

Cerebras just launched CS-4, up to 30x faster than GPUs on tokens per second per user, hitting over 4,400 on GPT-OSS-120B, roughly double the CS-3. Also had gpt 5.6 pro back into sols params from the cerebras speed numbers on this same chart and it landed around 1.5 trillion total 48 billion active,…

🚨 GPT-6 RUMORS ARE HEATING UP 👀🔥 OpenAI has "not officially confirmed GPT-6" or a launch in the coming weeks. But rumors suggest OpenAI's next model could bring: • A major generational shift beyond GPT-5.x • A significant leap in overall performance • Stronger reasoning and problem-solving •…

Our engineers barely open Slack, and that's where Viktor lives. Fixed that today. @viktor_com is now an OpenAI-compatible API and a hosted MCP server, in public beta. In the video I ask it for a branded snake game from opencode. 76 seconds later it's built and published, and it already knew our…

A few weeks after @jack built “Slack for AI agents,” … Salesforce announced Slack Code, where teams can put Claude, Codex, GitHub, Devin, and Vercel agents in one channel to assign tasks, track progress, review code, and open PRs.

Setup fx to use citadel (my model gateway) So I can use any local model or any of my existing subs without going through vercel this is how i use every harness btw > [@vercel_dev](https://twitter.com/vercel_dev): > Introducing fx, a tiny, open, native coding agent from Vercel Labs. > > Originally an…

1080P is where AI video starts showing off Seedance 2.5. Wan 3.0. MiniMax H3. Three heavy hitters. All showing off in HD. Which one caught your eye? Create in 1080P on WaveSpeed

New GPT-Image model in LM Arena right now. luna-lisa-alpha > [@chetaslua](https://twitter.com/chetaslua): > 🚨 New GPT-image is testing > > luna-lisa-alpha is being tested , this is new checkpoint last one was monalisa > > as always tested with my charlie kirk test for knowledge cut off > > > this…

Your entire media library, searchable by memory instead of filename, running right on your own machine. It’s a local-first AI file agent called MUZIM. Instead of managing folders, you work with what's actually inside your photos, videos, and documents. Here's how it comes together in a real…

Small change, big difference for Claude. Claude Code now offers a native 'Concise' output style. Turn it on in /config → Output style, or set "outputStyle": "Concise" in settings.json. Works great, and Boris says it's a band aid solution with more to come. > [@bcherny](https://twitter.com/bcherny):…

Grok Bot launched 9 days ago. Still so early. Crazy to imagine how much better it gets from here.

Slack launches Slack Code, introducing dedicated channels for real-time collaboration with AI coding agents. Tagging an AI agent like Anthropic’s Claude or Cognition’s Devin automatically spins up a project channel. Features include dedicated user tabs, live HTML previews, and code diff auditing…

Now building is a team sport. Introducing Slack Code: a new way for your team and AI agents, like @AnthropicAI's Claude Code, Claude Tag, Devin from @cognition, @github Copilot, @ChatGPT, and @vercel agents, to write, review, and ship code together. Just mention an agent to spin up a channel, pull…

Your job is your job Everything else is side quests Your family decides to visit next weekend and suddenly you’re finding hotels, restaurants, things to do, airport pickups and figuring out everyone’s schedule Her birthday is next week and you need a gift she’ll actually like....without accidentally…

Claude Opus-level coding AI is becoming something you can download and run on your own hardware. Ornith-1.5 is open-source, MIT-licensed, and its 397B MoE reportedly hits 86.1 Terminal-Bench 2.1 and 86 SWE-bench Verified. The proprietary moat is getting very thin. >…

Tencent's Hy4 model seems close to release In the Q2 investor presentation, Tencent said that HY4 will be a larger parameter model. According to a LatePost article, Hy4 will also be multimodal > [@MaxForAI](https://twitter.com/MaxForAI): > Breaking: Tencent has started gray-testing its brand-new…

Cursor agents can now watch your PRs and Slack threads, wake themselves up when something happens, fix CI, respond to feedback, and keep going until the job is done. You don’t have to keep prompting them. > [@cursor_ai](https://twitter.com/cursor_ai): > We're continuing to improve cloud agents in…

Qwen3.8-27B just got a proper cyber-focused uncensored & Run locally. Specifically tuned to answer cyber & offensive-security questions without the usual refusals. - Abliterated cyber-tuned - 100/100 on held-out cyber-offensive prompts - Scored 100 on their cyber-open eval. - Vision & MTP…

5M$ for one Domain 🙌🏻 > [@tryramp](https://twitter.com/tryramp): > We bought https://router.com. > > Now we're building the best one. > > Our customers buy quadrillions of tokens every month through Ramp, and we've run our own AI on this router for 3 years to keep costs down. > > Today it's yours…

Open-source video creation platform. Once deployed and configured with the API key, you can start selling right away. > [@zyailive](https://twitter.com/zyailive): > As promised, we kept our word and made Magiviz officially open source, with both frontend and backend released under the MIT license. >…

Most AI benchmarks ask whether a model can find the right answer. TRACES asks something harder: can the system reach a defensible conclusion when the answer isn’t known yet and show a reliable process for getting there? That’s a much more interesting way to evaluate scientific AI. @Apodex_AI >…

Cursor now supports the /goal command, along with new updates to cloud agents, Slack integration, and more! > Subagents can now run on their own virtual machines, each with an isolated copy of the project. > Cursor can now monitor your PRs, watch a Slack thread, or run scheduled tasks. >…

Stop Using the Same 200 Fonts as Everyone Else Every design tool hands you the same font list, which is why so much design ends up looking the same. Lovart just added a Font Generator that builds real working typefaces, either from an image you show it or from a sentence describing the vibe. I used…

Taimi-14B-Med v0.1 is here, laying the deployment foundation for medical Q&A, nursing guidance, health alerts, and service robots. 🩺 🤖 https://modelscope.ai/models/TMiRob/Taimi-14B-Med 🏆 77.4% on CMExam and 74.0% on CMB, reflecting the medical Q&A capability of the current base checkpoint. ⚡…

EdgeCrafter is just out on Hugging Face 🪶 detection, instance segmentation and pose from ONE compact ViT family of tiny models instead of three separate models (10M-50M params each!) ▶️ on Spaces https://huggingface.co/spaces/Intellindust/edgecrafter-detection

🚨 GPT-6 LEAK: WHAT SHOULD WE ACTUALLY BELIEVE? The rumors are getting harder to ignore... Over the past few weeks: ° OpenAI revealed Astra, as it's next major frontier model. ° An unreleased OpenAI model demonstrated surprisingly strong autonomous cyber capabilities during testing. ° OpenAI said…

A mysterious new image model checkpoint called “luna-lisa-alpha” has started appearing in public testing and early testers say it may be a successor to the previously spotted “monalisa” checkpoint. > [@chetaslua](https://twitter.com/chetaslua): > 🚨 New GPT-image is testing > > luna-lisa-alpha is…

🚨 GPT Image Update These are from luna-lisa , i will run prompts for you too comment what you wanna run Holy Shit look at this iphone fold UI mockup made by this beast , this will also improve the webdev coding > [@chetaslua](https://twitter.com/chetaslua): > 🚨 New GPT-image is testing > >…

It's interesting how Slack is developing. Most recently, there was the announcement of the deep integration of Claude into Slack (even Andrej Karpathy wrote about it), and now coding is being implemented. > [@Benioff](https://twitter.com/Benioff): > Don’t code alone. Slack Code is live. > Humans and…