Skip to content
K = 0.731
← All topics

On-device AI

AI that runs on phones, laptops and edge devices: small models, local inference and on-device chips.

16 picksLast 30 days 6 itemsCollected 66 items

New

Picks about On-device AI

10/14Wed
  1. Google DeepMind63

    Google DeepMind and Singapore GovTech deliver on-device Gemma model for scam detection

    Google DeepMind and Singapore’s Government Technology Agency (GovTech) have delivered a Gemma model fine-tuned on local scam conversations and optimized for typical smartphones in Singapore. A lightweight classifier screens incoming messages before the model evaluates broader conversation context and triggers warnings about suspected manipulation, with conversation processing kept on the device. Training excluded private or personally identifiable information, using anonymized datasets derived from scam reports and intelligence feeds.

    Why it matters: The two-stage detection pipeline offers a transferable approach to balancing smartphone power consumption with contextual scam analysis while keeping private conversations on the device.

10/8Thu
  1. NVIDIA Blog78

    NVIDIA and Microsoft announce RTX Spark and infrastructure for Windows agents

    At a Windows AI and Surface event in San Francisco, NVIDIA and Microsoft said they are designing software and hardware together for the agent era on Windows PCs. Microsoft announced that its OS-level Microsoft Execution Containers (MXC) infrastructure is generally available, allowing agents to run persistently in the background under system control.

    Why it matters: NVIDIA and Microsoft are delivering both the system-level containers and local compute needed to run agents, helping readers assess the conditions for local agent workflows.

10/7Wed
  1. Google DeepMind71

    Google DeepMind releases EmbeddingGemma 2, a lightweight open multimodal embedding model

    Google DeepMind released EmbeddingGemma 2, a 740M-parameter open model based on the Gemma 4 architecture and licensed under Apache 2.0. It maps text, images, audio, and video into a shared embedding space.

    Why it matters: EmbeddingGemma 2 maps text, images, audio, and video into one embedding space and reports concrete on-device memory and vector truncation figures, helping assess local multimodal retrieval.

10/2Fri
  1. NVIDIA Blog60

    NVIDIA DGX Spark adds 64GB version for scaling local AI with two-device clusters

    NVIDIA will release a DGX Spark with 64GB of unified memory this month through Acer, ASUS, Dell, Gigabyte, HP, and MSI. It goes on sale October 23, starting at $4,999, and retains the GB10 Grace Blackwell superchip, DGX OS, and full NVIDIA AI software stack. It supports running models with up to 100 billion parameters locally.

    Why it matters: The source gives the 64GB configuration's price, launch date, and two-device clustering option, helping readers assess the hardware threshold for running large models locally.

9/22Tue
  1. Hugging Face Blog64

    Transformers now supports GGUF-quantized models from llama.cpp

    Hugging Face has added GGUF support to Transformers. Users can select a GGUF checkpoint from the Hub and pass `gguf_file` to `from_pretrained` to generate locally, with no extra setup.

    Why it matters: Hugging Face's official post explains how to load GGUF directly in Transformers and shares comparison results, helping readers judge whether to move their local-inference workflow.

8/12Wed
  1. Google DeepMind71

    Google DeepMind releases SL2T sign-language-to-text model, first rolling out in Gboard and Live Transcribe on Pixel 11

    Google DeepMind has released SL2T, a multilingual sign-language-to-text model, bringing sign-language AI to consumer products for the first time. On Pixel 11, Gboard and Live Transcribe support dictation from American Sign Language (ASL) to English; more devices and languages are planned.

    Why it matters: The source describes the capabilities and limits of a multilingual sign-to-text model and where it is available, helping readers assess sign-language translation in real products.

7/9Thu
  1. Google Research62

    Google releases SensorFM, a large sensor foundation model for wearable-health data

    Google Research and Google DeepMind have introduced SensorFM, a large sensor foundation model that learns directly from unlabeled wearable data. Its pretraining corpus contains more than one trillion minutes of multimodal sensor signals from five million consenting participants across more than 100 countries, all 50 U.S. states, and over 20 Fitbit and Pixel Watch devices.

    Why it matters: The article reports pretraining-data scale, scaling experiments, and downstream-task comparisons, helping readers assess how generalizable wearable foundation models are.

7/8Wed
  1. Mistral AI71

    Mistral AI releases Robostral Navigate embodied-navigation model

    Mistral AI has released Robostral Navigate, its first embodied-navigation model. The 8B model uses a single standard RGB camera, with no LiDAR or depth sensor, and achieves a 76.6% success rate on R2R-CE validation unseen—9.7 percentage points above the best single-camera approach and 4.5 points above the best systems using depth or multiple cameras.

    Why it matters: The source reports navigation success using a monocular RGB camera and describes the training approach, giving a view of how sensor requirements for embodied navigation are changing.

6/27Sat
  1. Google Research60

    Google speeds up Gemini Nano on Pixel with frozen multi-token prediction

    Google Research has introduced an architecture that attaches a multi-token-prediction (MTP) head to the frozen Gemini Nano v3 model, speeding up on-device generation without changing the backbone weights.

    Why it matters: Google attaches an MTP head to frozen Gemini Nano v3 and reports concrete methods and gains under on-device memory and energy constraints.

6/9Tue
  1. Google DeepMind78

    Google DeepMind releases Gemma 4 12B, a unified encoder-free multimodal model

    Google DeepMind has released Gemma 4 12B, an agentic multimodal model aimed at laptops, positioned between the edge-focused E4B and the 26B MoE. It is the company's first medium-sized model with native audio input.

    Why it matters: Gemma 4 12B removes multimodal encoders and feeds visual and audio input directly into the LLM, giving readers a view of hardware requirements for local multimodal agents.

6/5Fri
  1. Google Research67

    Google Research studies passive heart-health monitoring with a phone camera

    In a paper published in Nature, Google Research describes PHRM, a system that uses the front-facing camera to record a few seconds of video after face unlock and estimate heart rate and resting heart rate in the background using deep learning.

    Why it matters: Google passively measures heart rate with a phone's front camera after unlock and has released its largest phone-video dataset to date along with a pretrained model.

4/3Fri
12/3Wed
  1. Mistral AI82

    Mistral releases Mistral 3 family, from 675B Mistral Large 3 to 3B, 8B, and 14B Ministral 3

    Mistral has released the Mistral 3 family: dense small models at 3B, 8B, and 14B, plus Mistral Large 3, a sparse MoE model with 41B active and 675B total parameters. All are open-source under Apache 2.0.

    Why it matters: Mistral has released Apache-2.0 weights ranging from 3B to 675B, giving readers a view of open-model options from edge devices to frontier scale.

3/17Mon
  1. Mistral AI71

    Mistral releases Mistral Small 3.1 with multimodal support and 128K context

    Mistral AI has released Mistral Small 3.1, improving text performance and multimodal understanding over Mistral Small 3. It extends the context window to 128K tokens, runs at 150 tokens per second, and is open-source under Apache 2.0.

    Why it matters: Mistral reports comparisons for a small model across text, multimodal, and long-context tasks, helping readers assess on-device deployment options.

1/30Thu
  1. Mistral AI78

    Mistral AI releases Mistral Small 3, an open 24B model under Apache 2.0

    Mistral AI has released Mistral Small 3, a 24B-parameter model optimized for low latency and open-sourced under Apache 2.0. It provides both pretrained and instruction-tuned checkpoints.

    Why it matters: A 24B model under Apache 2.0 is positioned against models three times its size, with quantization details for single-GPU local deployment to help readers assess its fit.