DeepMind Puts Singapore's Scam Detector on the Phone
DeepMind and Singapore's GovTech fine-tuned a lightweight Gemma to spot scams in conversations on-device, with no messages sent to the cloud. No performance figures have been published.
On-device AI News and Analysis
16 verified stories covering On-device AI, product updates and industry developments.
ModelBest open-sources MiniCPM-V 4.7, a 35B model in 3GB on a phone
ModelBest and OpenBMB release MiniCPM-V 4.7, including a 35B-A3B MoE that reportedly runs in about 3GB of memory on an iPhone 17 Pro Max at over 13 tokens/s.
Google Open-Sources 740M-Parameter Multimodal Embedding Model
Google open-sourced EmbeddingGemma 2, a 740M-parameter multimodal embedding model for on-device text, image, audio and video search, under Apache 2.0.
PrismML Squeezes 1-bit Model Onto Qualcomm's Glasses Chip
Qualcomm demoed PrismML's 1-bit, 2-billion-parameter Bonsai model running locally on Snapdragon AR1 Gen 1 smart glasses chips.
Hugging Face Open-Sources 207 Browser GPU Kernels
Hugging Face's new @huggingface/kernels loader ships 207 versioned WebGPU kernels on the Hub, posting a 2.57x geometric-mean speedup across 809 benchmark comparisons.
Apple's M6 Goes 2nm, M5 Ultra Packs 512GB for Desktop AI
Apple's M6 debuts on 2nm for the Mac mini, while the M5 Ultra brings 512GB of unified memory and 1.2TB/s of bandwidth to the Mac Studio for local AI inference.
Liquid AI Cuts Decoding Latency 3.18x With a 300M Draft Model
Liquid AI's LFM2.5-DSpark draft models bring speculative decoding to three on-device models, cutting decoding latency up to 3.18x with identical outputs.
Liquid AI revives Q4_0 quantization with 97% accuracy via distillation
Liquid AI's QAD training method restores Q4_0 GGUF checkpoints to 96-97% of BF16 accuracy, preserving the format's speed edge on Arm CPUs and older devices.
ModelBest starts A-share IPO tutoring
ModelBest has entered IPO tutoring with CITIC Securities, pushing China's edge-model race from open-source downloads into audited revenue, governance and deployment tests.
MiniCPM puts robot models on device
OpenBMB released MiniCPM-Robot with a 1.5B manipulation model, a 0.9B tracking model, and a deployment path focused on local latency.
PrismML squeezes a 27B model onto an iPhone
PrismML squeezes a 27B model onto an iPhone. A concise localization of the verified facts and the industry signal behind the story.
Perplexity Turns to Hybrid Inference With Intel
Perplexity’s Computex demo routes simple and sensitive agent work to local PCs while reserving complex reasoning for the cloud.
Chrome quietly put a 4GB Gemini Nano model on users' disks
Google says Gemini Nano supports on-device safety features in Chrome, but a quiet multi-gigabyte download has raised fresh questions about consent, privacy and the cost of browser AI.
Google switches Gemma 4 to Apache 2.0 license, releases four models
Google DeepMind released Gemma 4 on April 2, switching to the Apache 2.0 license and ranking third on the Arena AI open-source leaderboard with the 31B model.
Apple squeezes a 3B-parameter on-device AI model into the iPhone
Apple's AI strategy prioritizes on-device deployment and privacy over chasing the most powerful general-purpose models. The company's 3-billion-parameter Apple Foundation Models, optimized for Apple Silicon, run locally on the iPhone, while complex tasks are handled by its Private Cloud Compute.
Google brings NanoBanana image generation to devices
Google brings NanoBanana image generation to devices. The story explains the announcement, the strategic context and the practical risks for the people or companies affected.