https://brief-tech-news.com/tags/liquid-ai/Recent content in Liquid AI on BriefTechNews.comHugoenCopyright &copy;2021 <br> Designed &amp; Developed by [Gethugothemes](https://gethugothemes.com/)Fri, 16 Oct 2026 14:02:26 +0000https://brief-tech-news.com/blogs/2026-10-16-ai/Fri, 16 Oct 2026 14:02:26 +0000https://brief-tech-news.com/blogs/2026-10-16-ai/Meta released Muse Spark Open, an 816 B‑parameter multimodal model with a 1 M‑token context window and 42 B active weights, optimized for Apple M5 Ultra rigs via MLX and custom Helion kernels. Microsoft unveiled its MAI‑Base‑2 frontier pre‑training stack, a 2.8 T‑parameter MoE model that hit 21 % MFU on 16 000 GB200 GPUs in 38 days. Liquid AI announced d1‑ultra, a decision‑focused multimodal model priced at $0.042 / M tokens, beating Jev 1.14 on the Artificial Analysis Decision Index. Open‑weight models now account for 71.5 % of Vercel AI Gateway token volume but only 23 % of spend, reshaping cost strategies for production workloads.https://brief-tech-news.com/blogs/2026-10-14-ai/Wed, 14 Oct 2026 14:02:23 +0000https://brief-tech-news.com/blogs/2026-10-14-ai/OpenAI busted an employee with a hidden recorder, handed the case to the FBI, and confirmed no model assets were stolen. Google Cloud added Gemini 4 Argon, a long‑context, multi‑step model that pauses for human approval before irreversible actions. Anthropic unveiled “dynamic workflows” that let Claude agents self‑organize and write their own orchestration code, trimming human bottlenecks in complex pull‑request reviews. NVIDIA open‑sourced Run:ai, Physical Intelligence launched a Pi API for robot foundation models, and a slew of new models and tools - from Liquid AI’s 24‑billion‑parameter MoE to CopilotKit’s CLiF - hit the market.https://brief-tech-news.com/blogs/2026-10-06-ai/Tue, 06 Oct 2026 13:46:21 +0000https://brief-tech-news.com/blogs/2026-10-06-ai/OpenAI is courting a $30 billion round, with UAE funds and BlackRock poised to back up to $10 billion. The same day it announced EU‑required invisible text watermarks for ChatGPT and Codex. Instinct rolled out group‑chat AI assistants, while OpenAI’s new B2B Marketplace lets enterprises mix proprietary and open‑weight models under one bill. Elsewhere, Reflection unveiled a 501 billion‑parameter sparse MoE model, and Liquid AI’s d1 vision model promises 19‑200× cheaper inference. A slew of engineering‑focused releases - from Devin’s Git‑backed memory to Framefields’ GPU‑native video engine - round out the day.