The AI signal.
Minus the noise.

We’ve been playing around with Seedance 2.5 on CapCut and the AI Extend feature is probably our favorite part. You can keep extending the same video instead of making a completely new clip every time, which makes it much easier to build out an actual story. There’s an $80,000 cash rewards + Points…

Step by step, then all at once > [@SpaceXAI](https://twitter.com/SpaceXAI): > Grok 4.6 is live on Amazon Bedrock > > https://x.ai/news/grok-4-6-amazon-bedrock

we gave it one sentence: a group of friends traveling from qingdao to beijing. it looked up the trip itself and turned the idea into a full vlog. try it on Flowith.

OpenWolf optimizes context management and token utilization for AI coding assistants using seven invisible lifecycle hooks. https://github.com/cytostack/openwolf

Imagine Computer is here. It remembers you, works for you, and builds and grows WHATEVER you want. Now yours👇 https://www.imagine.art/computer

This is actually insane.... open source models that self improve end to end (generate their own tasks → scaffolds → RL rollouts) and now go toe to toe with Claude Opus 4.8 on agentic + coding benchmarks… all under MIT with phone ready 9B and quantized versions dropping today the open source…

Google Photos might add map view to Ask Photos search https://9to5google.com/2026/08/19/google-photos-ask-photos-map/ by @chaosromero

Gemma 4 is absolutely flying on @SarvamForDevs inference. 260 TPS streaming vs 148 TPS for GLM-5.2, with a 0.5s TTFT vs 17.1s. 1.8× faster in streaming and starts generating 33× sooner.

We’re releasing new Qwen3.8-27B GGUFs with 10% higher accuracy. Unsloth Dynamic V3 outperforms others by >10% on Div-300, KLD & more benchmarks. We also release 1-bit quants that retain 77% accuracy. Run on 8GB RAM. Blog: https://unsloth.ai/docs/basics/dynamic-3.0-ggufs GGUF:…

We’re releasing new Qwen3.8-27B GGUFs with 10% higher accuracy. Unsloth Dynamic V3 outperforms others by >10% on Div-300, KLD & more benchmarks. We also release 1-bit quants that retain 77% accuracy. Run on 8GB RAM. Blog: https://unsloth.ai/docs/basics/dynamic-3.0-ggufs GGUF:…

LTX 2.5 📸ReelBids dolly-in camera LoRA (V2) v2 improved than earlier V1 with 1250 and 1500 steps drives a dolly-in camera move at seven selectable speeds, 👇 https://huggingface.co/AhsanHareem/reelbids-ltx25-camera-lora-v2

LTX 2.5 📸ReelBids dolly-in camera LoRA drives a dolly-in camera move at seven selectable speeds, for LTX2.5 image-to-video -one adapter covers the whole range — unlike the official LTX-2 19b camera LoRAs, which are single-speed and 19b-only. 👇…

Wow!! > [@Replit](https://twitter.com/Replit): > Replit Free Mode, powered by @OpenAI GPT-5.6 Luna. > > Let’s make intelligence accessible to everyone.

🔥Big breaking news! Open-source AI is blowing up the scene again! The latest creation from the U.S. DeepReinforce team, Ornith-1.5, is officially out! Performance is right up there with Claude Opus 4.8, and it’s fully open source under MIT! Key highlights you have to see: 1️⃣ Three models fully…

Lightx2V's MiniMax-H3🙂Prompt Rewriter LoRA 8B adapter -turns short user requests into production-oriented prompts H3 joint audio-video generation -supports text-only and keyframe-conditioned rewriting for T2VA, T2AV, I2VA, I2AV, L2VA, L2AV,FL2VA 👇…

ICYMI: Gemini 3.7 Flash 🤝Google Search… and it’s 🔥🔥🔥-- big improvements in instruction following & accuracy > [@rmstein](https://twitter.com/rmstein): > News Flash⚡ We’re bringing Gemini 3.7 Flash to Search! > > ->Better at following instructions + understanding your intent, so you get even more…

ICYMI: Gemini 3.7 Flash 🤝Google Search… and it’s 🔥🔥🔥-- big improvements in instruction following & accuracy > [@rmstein](https://twitter.com/rmstein): > News Flash⚡ We’re bringing Gemini 3.7 Flash to Search! > > ->Better at following instructions + understanding your intent, so you get even more…

🔥 Thank you, Liquid AI, for this new quant method. Unsloth, Google, and Nvidia aren't the only ones pushing smaller quantizations. Liquid AI released new Q4_0 GGUFs that were trained to survive 4-bit quantization. And they're retaining roughly 97% of their BF16 performance. Liquid calls the…

Prompting each video clip is (largely) dead. Welcome to the era of vibe-directing 🎬 > [@yapper_so](https://twitter.com/yapper_so): > Vibe-video making is here with agents > > Yapper does for videos what Lovable did for websites. > > Just say your idea and Yapper’s agent makes a full-length video…

Grok 4.6 now on Bedrock > [@SpaceXAI](https://twitter.com/SpaceXAI): > Grok 4.6 is live on Amazon Bedrock > > https://x.ai/news/grok-4-6-amazon-bedrock

MiniMax H3 just went unlimited on Runway. Generate responsibly. Or don't. It's unlimited 😌 > [@runwayml](https://twitter.com/runwayml): > MiniMax H3 is now unlimited on Runway. > > For a limited time, Max plan users can generate as much as they want with no caps, no counting, just unlimited access…

Holy moly open source grok bot? https://github.com/milind-soni/OpenMausBot

✨ Doubao Seed 2.1 Pro — now 20% OFF on ZenMux. Give it one rough brief. It'll come back with a story, a hook, and three better angles. Meet Doubao Seed 2.1 Pro, the Idea Launcher. 💡 Strong fit for writing, presentations, multimodal content, and creative workflows 👉 https://zenmux.ai/ai-workmates…

a new open-weight model family where the model card ships with Hermes Agent setup out of the box. point your Hermes Agent at a local Ornith 1.5 server, 9B to 397B! benchmark coming soon :) > [@ornith_](https://twitter.com/ornith_): > Aloha! 🌺Introducing Ornith-1.5, a family of open-source LLMs…

In The Matrix, it becomes obvious it’s a simulation when the black cat appears twice. World models fail similarly: visually the generated simulation looks realistic, but the underlying distribution is wrong. Introducing CaliBench, a measurement of the randomness of reality! >…

In The Matrix, it becomes obvious it’s a simulation when the black cat appears twice. World models fail similarly: visually the generated simulation looks realistic, but the underlying distribution is wrong. Introducing CaliBench, a measurement of the randomness of reality! >…

Ornith just launched Ornith-1.5 today. Here's what you need to know. Ornith-1.5 is a family of open-source LLMs in three sizes: 9B Dense, 35B MoE, and 397B MoE, trained using self-improving strategies. The core idea is a self-improvement loop: the model proposes its own tasks, builds task-specific…

One story opening. Thousands of possible endings. CapCut is challenging creators to build a 3+ minute continuation using Seedance 2.5, AI Video, AI Edit and AI Extend. The most creative execution could take home $ 50K. > [@capcutapp](https://twitter.com/capcutapp): > Still haven’t joined yet? > >…

Grok 4.6 is now available on Amazon Bedrock > [@SpaceXAI](https://twitter.com/SpaceXAI): > Grok 4.6 is live on Amazon Bedrock > > https://x.ai/news/grok-4-6-amazon-bedrock

Grok 4.6 is live on Amazon Bedrock https://x.ai/news/grok-4-6-amazon-bedrock

Sonic-3.6 just took #1 on both TTS leaderboards — provider voices and controlled voice streaming. 44 languages. Three months after 3.5. The gap between this and everything else is not close.

🚨Ornith 1.5 looks absolutely ridiculous A 397B open model with roughly Opus 4.8 level performance across coding, agents and reasoning • Terminal Bench 2.1: 86.1 • SWE Bench Verified: 86.0 • SWE Bench Pro: 65.1 • DeepSWE: 56.0 • HLE: 44.6 • ClawEval: 81.4 I can't believe a 397B model actually beats…

HY3 from @TencentHunyuan will remain free via Nous Portal through the end of the month > [@yeahfortommy](https://twitter.com/yeahfortommy): > One more time by popular demand, HY3 free through the end of August > > enjoy your tokens

Replit just launched Free Mode today, powered by OpenAI's GPT-5.6 Luna. Here's what you need to know. Free Mode lets Replit's Core and Pro subscribers build and chat with the Agent for free, using GPT-5.6 Luna as the model behind it. It's not a new free tier, you still need a paid subscription, but…

Ornith has released the Ornith-1.5 series of models. Its standout feature is self-improvement: the model no longer relies on manually fixed datasets, but instead autonomously builds a closed loop of “propose new tasks -> generate task-specific scaffolding/evaluation environments -> generate solution…

seedance 2.5 真的是流量密码啊！ 日女版，是熟女的感觉！ 👉 https://evolink.ai/ai-video-generator/seedance?utm_source=twitter&utm_medium=thread&utm_campaign=seedance25-18 Main subject: young Janpanese woman, early 20s, polished early-2000s Japanese everyday appearance, fitted cream-colored fine-knit cardigan layered over…

I had my first magic moment with Grok Bot yesterday (coming off of a week of vacation and reading about the vibes everybody was sending). I couldn't agree with you more. This is the most human-like agent experience I've had so far. The simplicity of how they're serving it up is really the key, I…

Do you keep opening Grok Bot and then just sitting there? This is for that moment: https://botdirectory.ai > [@elie2222](https://twitter.com/elie2222): > Introducing Bot Directory, Grok Bots you can set up with a single prompt: > https://botdirectory.ai > > X has been full of amazing bot setups.…

🔥How do you put Claude Code’s gorgeous UI into DeepSeek Harness?! Make DeepSeek Harness’s UI as clean and comfortable as Claude Code’s. DeepSeek Code combines the two perfectly! Core highlights: 🔹 Deep compatibility with all DeepSeek Harness plugin features and native upgrades, with the ecosystem…

Shiru manages anime libraries with direct file playback and no transcoding. https://github.com/RockinChaos/Shiru

Honestly, CapCut putting Seedance 2.5, AI Edit, and AI Extend into the same workflow makes a lot of sense. You can generate a scene, edit it with prompts, then keep extending the story without constantly jumping between tools. And there’s $80K up for grabs in the Seedance 2.5 Wave 2 challenge >…

GPT-5.6 Sol is 50% off on OpenCode Zen until September 18th

New Ornith models ✨ The most interesting for me is the 35B MoE since there is no Qwen3.8-35B. Will test it out soon. > [@ornith_](https://twitter.com/ornith_): > Aloha! 🌺Introducing Ornith-1.5, a family of open-source LLMs spanning 9B Dense, 35B MoE, and 397B MoE, trained with self-improving…

GitLab Duo CLI brings agentic AI directly into your terminal. From troubleshooting pipelines and planning changes to custom workflows, headless automation and governance controls. Here are 7 things you can do with Duo CLI.

Open-source models are entering a new phase. Ornith-1.5 is not just increasing parameters; it's changing how models learn. It uses a self-improving loop where the model generates tasks, creates frameworks, produces solutions, and learns from these through reinforcement learning. This represents a…

Block, led by Twitter founder Jack Dorsey, has released an AI agent app called “Berd” (probably macOS only) that manages multiple AI agents as characters and lets you assign them a variety of tasks. https://gigazine.net/news/20260819-berd-ai-agent/

🚨 Ornith just released Ornith-1.5, a new open-source model, and its putting up some seriously strong numbers. It reportedly matches or beats Claude Opus 4.8 across several coding, reasoning and agentic benchmarks: • 86.1 Terminal-Bench 2.1 • 86.0 SWE-bench Verified • 56.0 DeepSWE • 44.6 HLE • 71.2…

Anyone tried the Ornith 1.5 397B model? Thinking of giving it a shot today. For science.

Beats Fable? Wait, maybe i give THEM my money now. > [@cline](https://twitter.com/cline): > GLM-5.3 has released, beating Fable & the new DeepSeek V4-Pro 0813 from just yesterday on Terminal-Bench. > > It used GLM-5.2 as the base model with gains from more compute spent than ever before on…

Anyone tried the Ornith 1.5 397B model? Thinking of giving it a shot today. For science. https://huggingface.co/ornith-ai/Ornith-1.5-397B-NVFP4

🔥 GLM-5.3 🆚 DeepSeek V4 Pro 0813, maxed-out real-world testing across multiple scenarios! Benchmarking against Fable5! 💻 Complex Coding / Long-horizon Agent tasks: GLM goes straight to true Fable5 level! ⚡ Speed & cost-effectiveness: GLM5.3 is lightning fast, with visibly strong value in token…

Video editing just got a major upgrade. ⚡ With Gemini Omni in Google Vids, you can edit any footage — from phone video to AI clips — using simple text prompts. Make your rough cuts look professional-grade, or tweak scenes without a reshoot. Learn more in our latest Workspace Drop →…

Looks like I finally got access to GLM-5.2 and Gemma 4 on @SarvamAI inference. Let’s see how they perform on Sarvam’s inference 👀

Orthith-1.5 Opensource Model was just released. With strong benchmark scores with SWE-Bench and GPQA Diamond. Impressive little model. > [@ornith_](https://twitter.com/ornith_): > Aloha! 🌺Introducing Ornith-1.5, a family of open-source LLMs spanning 9B Dense, 35B MoE, and 397B MoE, trained with…

AI is getting out of hand! These shots are 100% AI-made. Created with Seedance 2.5.

Very interesting model created by self-improvement thought self-generated tasks. > [@ornith_](https://twitter.com/ornith_): > Aloha! 🌺Introducing Ornith-1.5, a family of open-source LLMs spanning 9B Dense, 35B MoE, and 397B MoE, trained with self-improving strategies. > > It achieves…

Great news for developers using Q4_0 GGUFs: We just released new 4-bit ckpts trained with Quantization-Aware Distillation (QAD). QAD a technique to distill a high-precision teacher model into a quantized student model. This allows us to recover most of the accuracy lost to quantization (landing at…

🔥 each::labs Video API is live. The model gives you the clip. Production wants the other half. → Reframe to 9:16. Burn captions. Match color. Duck the music. → Normalize loudness. Cut silence. Pull a thumbnail. Package for the web. → Cut without re-encoding. That is nine of the forty. That half now…

Looks really good, this will make scanning houses way faster cheaper for real-estate listings! > [@SpAItial_AI](https://twitter.com/SpAItial_AI): > Large, consistent, and high-quality 3D worlds from simple iPhone recordings. > > Works for multi-room scenes, fills missing gaps consistently while…

WAN 3.0 proves it can deliver Seedance 2.5 level quality while keeping the comedic timing intact 😂#Topview #aivideo #wan3