LFM2.5-VL-3B does screen grounding on a phone.

Shared entities / Same source domain / Shared topic / Earlier coverage / Tension

Two Labs Ship Sub-500MB Agentic Models in the Same Week. Edge AI Agents Just Became Real.

Both cover LFM2, Liquid AI; reported by the same outlet (liquid.ai); overlapping topics (agent, calling).

Shared entities / Shared topic / Earlier coverage

Liquid AI's LFM2.5-2.6B clocks 260 tok/s on a 3090, and one practitioner named the exact job it's for.

Both cover LFM2, Liquid AI, M5 Max; overlapping topics (calling, desktop); earlier LFM2 coverage from 2026-08-09.

Shared entities / Earlier coverage / Tension

DGX Spark crushes Strix Halo on prompt processing (1,723 vs 339 tok/s) but token generation is nearly even.

Both cover M5 Max, Ryzen AI Max; earlier M5 Max coverage from 2026-06-28; pushes against this story (but).

Shared entities / Earlier coverage

Meta ships Muse Glimmer, a 30B open agent model

Both cover M5 Max, Ryzen AI Max; earlier M5 Max coverage from 2026-08-10.

Liquid AI's LFM2.5-2.6B beats an 8B Gemma on tool use in under 2.5 GB.

Both cover LFM2, Liquid AI; earlier LFM2 coverage from 2026-08-06.

Shared entity: Screen / Shared topic / Earlier coverage

"Screen recording as prompt" is a contested category now, with at least five entrants.

Both cover Screen; overlapping topics (agent, screen); earlier Screen coverage from 2026-08-06.

Shared entity: VLM / Shared topic / Earlier coverage

ExtractBench finds VLM agents truncate long record lists

Both cover VLM; overlapping topics (agent, grounding); earlier VLM coverage from 2026-08-03.

Shared entity: ScreenSpot / Shared topic / Earlier coverage

trycua/cua is positioning as the open "Computer-Use 2.0" stack.

Both cover ScreenSpot; overlapping topics (agent, computer-use); earlier ScreenSpot coverage from 2026-07-17.