Marlin-2B: a tiny VLM to extract structured information from videos (www.reddit.com)
model roundup
Gemini 2.5
-
Hi all! Shubham and Aryan here, putting out our first open source VLM release built on top of Qwen3.5-VL Story time: we were building video editing agents for social-media content and were using Gemini-2.5-Flash to analyse IG reels and fin…
-
Google I/O is tomorrow. What are we expecting? (www.reddit.com)
I think the only confirmed/leaked feature is Gemini Omni, which is some sort of video model, but it's not really clear to me if that's a new video model or just another form of Veo. It also seems a new Gemini Flash model (3.2?) is likely.