The AI signal.
Minus the noise.

We're on a roll at @liquidai 🥳 Our latest release LFM2.5-VL-3B is out. It's a lightweight, vision-language model that reads screens, documents, and the physical world. Congrats to our amazing VLM colleagues and my favorite deskmates when I visit our SF office! @sina_semnani

Grok 4.6 just released and it’s already a massive jump It outperformed Fable 5 Max and GPT-5.6 Sol Max on GDPVal-AA v2 benchmark https://t.co/Xw159JLj5z

👁️ LFM2.5-VL-3B is here! it comes with new/improved vision-capabilities: • screen understanding • better OCR • better grounding • tool use check it out on hugging face: https://t.co/Y4Elf3kC6U https://t.co/kmnmhhcEw4

WeChat released its models: 2 versions, WeLM-80B, with 80B total parameters and 3B active parameters; WeLM-617B, with 617B total parameters and 23B active parameters;

2.0 vs 2.5 — which one's your pick? Seedance 2.5 brings text, image, video, and audio all as reference. More control. More consistency. From start to finish. https://t.co/CpIbcQC2oC

CLAUDE OPUS 5, FABLE 5 AND GPT-5.6 SOL ARE FREE ON SIGNUP, NO CARD • Token Harbor > Claude Opus 5, Claude Fable 5, GPT-5.6 Sol, Kimi K3, MiMo v2.5 and DeepSeek V4 Flash on one key: https://t.co/c9PZZLAY2R > Sign up with Gmail, create the API key, credits land straight away. >…

🚨 CHINA'S AI MODEL WAR IS MOVING INTO APPS. Weixin just introduced WeLM, a new MoE model family: • WeLM-80B: 80B total parameters, only 3B active, already powering Xiaowei, Weixin's native AI assistant. • WeLM-617B: 617B total / 23B active: being developed for harder https://t.co/7ZzwlJGSdd

Grok 4.6 is coming today........

Topview @TopviewAIhq just launched a pretty nice 30s Challenge with a solid cash prize. There are 8 winners in total and the Gold Award winner gets $10,000. Create a 30 second AI video with Topview × Seedance 2.5, share it on X, TikTok, Instagram or YouTube and submit it https://t.co/34YmvjEgyr

Flux 3 is free to use until Monday (Beijing time). It still lags behind Seedance, but it can still hold its own against Minimax H3. Use it here: https://t.co/O1J9N5jm6e https://t.co/LNj62ja8Cl

Mi chino no es fluido, pero se supone que en 4 horas salen los pesos de Qwen 3.8 https://t.co/dMgTYKZOLP https://t.co/npiooov9kQ

What if your alarm clock didn’t just wake you up… but literally DRAGGED you out of bed? 😂⏰ I brought this crazy storyboard to life with Seedance 2.5 the character consistency, motion, expressions, and comedic timing are AI video is getting seriously wild. https://t.co/NXAEKAaLBn

[OpenSource - Video Transcription & AI Summarization] Sitting through a 2-hour podcast or video just to extract a few key insights wastes hours of valuable time. 🛑 AI-Video-Transcriber (3.1K+ GitHub stars) by wendy7756 is an open-source tool that turns raw video links into…

We built something new on QwenCloud: QwenCloud Arena — an Arena for agents solving real-industry challenges! We're taking Agents into industries and real business problems. Every task comes from real business scenarios, because the Agents worth building are the ones that https://t.co/bG9UB9nvyr

I’m extremely excited about OpenAI’s new Astra model, especially because people who’ve had access to it told me it’s absolutely incredible. But the thing I’m most impatient to get my hands on is 750 tokens per second on Cerebras chips. That’s going to be completely insane and

LTX 2.5 | 1920×1080 [24f] | Generated a 15-second video in just 4 minutes with Rtx 5080[16gb Vram] quality is surprisingly good. It’s faster than MiniMax and produces higher-resolution output, but MiniMax still seems better in terms of overall quality and prompt adherence. I https://t.co/OTlxY7bDTL

OpenAI's new cybersecurity model completes 95% of exploit-development prompts that its own flagship model refuses. That number is everywhere this week. What's not getting repeated: on OpenAI's own vulnerability report writing eval, that same model scores worse than the flagship

Xiucai webhook deployment insertion test on 2026-08-12.

OpenAI finally launched a ChatGPT desktop app for Linux. ChatGPT, Work, and Codex into one interface. Codex already worked on Linux through its CLI, so the new app mainly adds the integrated desktop layer. The preview supports Ubuntu 24.04 and 26.04 LTS, Debian 13, and Fedora https://t.co/XCzJrzHoQw

𝐒𝐞𝐞𝐝𝐚𝐧𝐜𝐞 𝟐.𝟓 upgrades lighting and detail that holds up frame by frame. Made with Seedance 2.5 on Loova. ⌛️Limited offer: 60% OFF