On-Device AI
10 pieces on On-Device AI.
News & Analysis
Architectural Specificity Now Outperforms GPU Scaling Alone
Three independent data points from one week show that workload-specific architecture delivers gains raw GPU procurement cannot match at any price.
Google's Feb 2027 Android Memory Rules Target On-Device AI Apps
Google mandates new dynamic memory and bitmap thresholds for Android apps by February 2027, driven by AI-related chip shortages hitting low-end devices hardest.
Liquid AI's Pipette Benchmarks 1,000+ On-Device Configs, Not Just Models
Pipette measures model + quantization + runtime + device together, exposing gaps like 78.4% vs 33.8% throughput retention between two 350M models.
WhatsApp Scam Alert Runs ML On-Device with Confidential VM Telemetry
WhatsApp's beta Scam Alert feature classifies messages locally, routes telemetry through confidential VMs with differential privacy, and verifies models via a third-party transparency ledger.
Software Extraction Beats Hardware Acquisition at the AI Frontier
GLM-5.3, Kog, and Needle 2 show that post-training, bandwidth recovery, and compression outperform new compute spend in August 2026.
Google DeepMind SL2T Brings Sign Language Dictation to Pixel 11
SL2T scores 70 BLEURT zero-shot on FLEURS-ASL, ships free in Gboard and Live Transcribe on Pixel 11 across 50+ sign languages.
Google DeepMind SL2T Brings Sign Language Translation to Pixel 11
SL2T scores 70 BLEURT zero-shot on FLEURS-ASL, ships in Gboard and Live Transcribe on Pixel 11, trained on 100,000+ hours across 50+ sign languages.
Liquid AI LFM2.5-VL-3B: 3.1B On-Device Vision-Language Model
Liquid AI's 3.1B-parameter LFM2.5-VL-3B scores 69.4 across 28 vision benchmarks, matches 4.7B rivals, and adds tool calling for on-device agents.
Google DeepMind SL2T: Sign Language to Text at Consumer Scale
SL2T scores 70 BLEURT zero-shot on FLEURS-ASL, powering Gboard and Live Transcribe on Pixel 11 with ASL-to-English translation.
Meta Muse Glimmer: 30B Open-Weight On-Device Agent Model
Meta releases Muse Glimmer, a 30B open-weight model under Apache 2.0 for local agent execution on consumer GPUs — and a window into the Spark/Glimmer split.