AI Dynamics

Global AI News Aggregator

@maximelabonne

Real-time video captioning in browser with LFM2-VL WebGPU

By

@maximelabonne

–

13 March 2026 18h46

Real-time video captioning in your browser, powered by LFM2-VL

Another exceptional demo by the great @xenovacom! https://t.co/KRvDEpj138
— Maxime Labonne (@maximelabonne) 13 mars 2026

Real-time video captioning in your browser, powered by LFM2-VL Another exceptional demo by the great @xenovacom! Xenova (@xenovacom) Real-time video captioning in your browser with @LiquidAI's LFM2-VL model on WebGPU. Sending every frame to a server was never going to be the answer. Imagine the bandwidth, latency and cost. Local inference. No server costs. Infinitely scalable. This is the way. — https://nitter.net/xenovacom/status/2032504624024854673#m

→ View original post on X — @maximelabonne, 2026-03-13 17:46 UTC

13 March 2026
Introduction to LLM Post-Training Techniques Presentation

By

@maximelabonne

–

06 March 2026 11h46

A big thank you to @itsmaddox_j for the invitation, and to professors @niclane7 and @contactrika for having me! slideshare.net/slideshow/int…

→ View original post on X — @maximelabonne, 2026-03-06 10:46 UTC

6 March 2026
53 Slides on Post-Training Algorithms and Data Quality

By

@maximelabonne

–

06 March 2026 11h45

I'm releasing 53 slides on post-training, covering core algorithms like DPO and GRPO, as well as data quality, synthetic data pipelines, and on-policy training. I had the pleasure of presenting it yesterday as a guest lecturer in Cambridge, UK

→ View original post on X — @maximelabonne, 2026-03-06 10:45 UTC

6 March 2026
LocalCowork: AI Agent Running on MacBook with 67 Tools

By

@maximelabonne

–

05 March 2026 16h55

> 385ms average tool selection. > 67 tools across 13 MCP servers. > 14.5GB memory footprint. > Zero network calls. LocalCowork is an AI agent that runs on a MacBook. Open source. 🧵

→ View original post on X — @maximelabonne, 2026-03-05 15:55 UTC

5 March 2026
1.2B LLM runs at 200 tokens per second in browser

By

@maximelabonne

–

25 February 2026 19h47

1B model running over 200 tok/s in your browser 👀 https://t.co/JArzxn7FN8
— Maxime Labonne (@maximelabonne) 25 février 2026

1B model running over 200 tok/s in your browser 👀 Xenova (@xenovacom) Okay, this is actually insane… You can now run LFM2.5-1.2B-Thinking (a 1.2B parameter LLM from @LiquidAI) at over 200 tokens per second directly in your browser on WebGPU! 🤯 Zero install. Fully private. Blazingly fast. Powered by Transformers.js and ONNX Runtime Web — https://nitter.net/xenovacom/status/2026727703836004796#m

→ View original post on X — @maximelabonne, 2026-02-25 18:47 UTC

25 February 2026
Introducing the LFM2-24B-A2B Model on Hugging Face

By

@maximelabonne

–

24 February 2026 18h48

huggingface.co/LiquidAI/LFM2… [Translated from EN to English]

→ View original post on X — @maximelabonne, 2026-02-24 17:48 UTC

24 February 2026
Unfortunate timing: 24B-A2B release coincides with Qwen3.5 launch

By

@maximelabonne

–

24 February 2026 18h01

Releasing a 24B-A2B model on the same day as Qwen3.5-35B-A3B is NOT great timing 🥲 Qwen (@Alibaba_Qwen) 🚀 Introducing the Qwen 3.5 Medium Model Series Qwen3.5-Flash · Qwen3.5-35B-A3B · Qwen3.5-122B-A10B · Qwen3.5-27B ✨ More intelligence, less compute. • Qwen3.5-35B-A3B now surpasses Qwen3-235B-A22B-2507 and Qwen3-VL-235B-A22B — a reminder that better architecture, data quality, and RL can move intelligence forward, not just bigger parameter counts. • Qwen3.5-122B-A10B and 27B continue narrowing the gap between medium-sized and frontier models — especially in more complex agent scenarios. • Qwen3.5-Flash is the hosted production version aligned with 35B-A3B, featuring: – 1M context length by default – Official built-in tools 🔗 Hugging Face: huggingface.co/collections/Q… 🔗 ModelScope: modelscope.cn/collections/Qw… 🔗 Qwen3.5-Flash API: modelstudio.console.alibabac… Try in Qwen Chat 👇 Flash: chat.qwen.ai/?models=qwen3.5… 27B: chat.qwen.ai/?models=qwen3.5… 35B-A3B: chat.qwen.ai/?models=qwen3.5… 122B-A10B: chat.qwen.ai/?models=qwen3.5… Would love to hear what you build with it. — https://nitter.net/Alibaba_Qwen/status/2026339351530188939#m

→ View original post on X — @maximelabonne, 2026-02-24 17:01 UTC

24 February 2026
LFM2-24B-A2B: Liquid AI’s Fastest 24B MoE Model Released

By

@maximelabonne

–

24 February 2026 16h44

Meet LFM2-24B-A2B, @liquidai's largest model. 24B MoE, 2B active. Blazing fast even on CPU. Available now in LM Studio 👾💧 lmstudio.ai/models/lfm2-24b-…

→ View original post on X — @maximelabonne, 2026-02-24 15:44 UTC

24 February 2026
Liquid AI releases LFM2-24B-A2B model for on-device inference

By

@maximelabonne

–

24 February 2026 15h36

ollama run lfm2:24b-a2b .@liquidai's latest on-device model is here! It's the largest LFM2 model yet, and is designed to run fast on device, and fits on devices with 32GB of unified memory. Liquid AI (@liquidai) Today, we release our largest LFM2 model: LFM2-24B-A2B 🐘 > 24B total parameters > 2.3B active per token > Built on our hybrid, hardware-aware LFM2 architecture It combines LFM2’s fast, memory-efficient design with a Mixture of Experts setup, so only 2.3B parameters activate each run. The result: best-in-class efficiency, fast edge inference, and predictable log-linear scaling all in a 32GB, 2B-active MoE footprint. 🧵 — https://nitter.net/liquidai/status/2026301771539202269#m

→ View original post on X — @maximelabonne, 2026-02-24 14:36 UTC

24 February 2026
Liquid AI Releases Largest LFM2-24B-A2B Model with Fast Inference

By

@maximelabonne

–

24 February 2026 15h31

Early checkpoint of our biggest LFM2 model to date 🎉 It shows good scaling and extremely fast inference vs. gpt-oss-20b and Qwen3-30B-A3B We'll release an LFM2.5 version with more pre-training and RL in a few months Liquid AI (@liquidai) Today, we release our largest LFM2 model: LFM2-24B-A2B 🐘 > 24B total parameters > 2.3B active per token > Built on our hybrid, hardware-aware LFM2 architecture It combines LFM2’s fast, memory-efficient design with a Mixture of Experts setup, so only 2.3B parameters activate each run. The result: best-in-class efficiency, fast edge inference, and predictable log-linear scaling all in a 32GB, 2B-active MoE footprint. 🧵 — https://nitter.net/liquidai/status/2026301771539202269#m

→ View original post on X — @maximelabonne, 2026-02-24 14:31 UTC

24 February 2026