AI Briefing#edge-ai
The latest AI and developer news about #edge-ai, with the original source and a short summary.
The latest AI and developer news about #edge-ai, with the original source and a short summary.
Optimization reduces planning time from 0.6–0.8 s to 45 ms, enabling 20 Hz camera processing on a single board.
The optimized model processes input from 6 cameras and recent motion data using a reinforcement learning training loop.
The new backend executes torch.export .pte files directly in Chrome, Safari, and other WebGPU-enabled browsers without format conversion.
The second-generation accelerator supports 20B to 100B parameter models at 10-30 tokens per second, with projected 6.7x performance gains via Samsung LPDDR5x-PIM.
Built on llama.cpp, the tool automatically allocates model weights to VRAM, RAM, or flash storage to maximize prefill and decode speeds on heterogeneous hardware.
Cactus Compute's Mirage model uses 51.3M parameters to perform multi-task visual analysis on mobile NPUs, enabling text-only LLMs to answer visual questions without fine-tuning.
NVIDIA announced that Nemotron 3.5 Lightning generation speed on DGX Spark is up to 1.73 times faster when DSpark is applied.
The 4GB model targets voice assistants and robotics on RTX, Jetson, and DGX Spark hardware.
Liquid AI has released six 2B-scale LFM3 models supporting execution on edge devices and multimodal processing.
The inference stack targets 10+ backends including CPUs, GPUs, and NPUs, with most components written in Rust.
The research demonstrated a model running on 15 Android TV boxes at 0.39 tokens per second and agent-driven deployment success rates of up to 84%.
NVIDIA released Nemotron 3.5 Super VL, a 120-billion-parameter open model, alongside two edge-optimized Bolt models for local agentic workloads.
The new framework allows a single model to span 1, 2, 3, or 4 billion parameters, outperforming dense counterparts by 2-5% on knowledge-intensive benchmarks.
Neuphonic has open sourced NeuDecide, a 43 MB audio-to-tool model that achieves 72.4% tool accuracy on SLURP benchmarks without transcription, running efficiently on a single CPU thread.
The new d1-3B model achieves a score of 48.57 on the Decision Index 0.2.1, outperforming larger models like Decider 35B-A3B.
The open-source engine delivers 24 kHz audio with two English voices, scoring 4.46 on UTMOS naturalness evaluations.
The model achieves an 11.1 ms time-to-first-token on an Apple M4 Pro CPU and supports seven languages.
The new module supports PCIe cascading and runs on embedded Linux and Android systems.
Gemma 4 offers five model sizes under Apache 2.0 license, enabling startups to reduce latency and infrastructure costs compared to relying solely on frontier APIs.
The new model uses Glow-TTS architecture and targets edge devices and CPU inference.
The new 280M-parameter draft model accelerates LFM2.5-VL-3B inference by up to 3.13x on device and 2.66x on H100 GPUs.
Next GenAi has released 'Rebekah 1.8B', an open-source LLM for edge deployment focused on data curation.
PrismML has released 'Ternary Bonsai 2 27B', a 27B multimodal LLM quantized to ternary values that maintains 98.2% of FP16 performance.
Cactus Compute has released 'Needle 3', an ultra-compact foundation model ranging from 29MB to 121MB, stating that its 4-layer subnetwork achieves automation performance equivalent to DeepSeek V4 Flash.