AI Dev and Latest Releases
[Open Source AI] NVIDIA AI Open-Sources ViPE (Video Pose Engine): A Powerful and Versatile 3D Video Annotation Tool for Spatial AI. ViPE turns raw “in-the-wild” video into metric 3D geometry by combining classical SLAM/SfM methods with deep learning. It estimates camera intrinsics, camera pose, and dense depth maps; masks out moving objects; works with various camera types (fisheye, 360°, etc.); and is reasonably fast (3-5 FPS on a GPU). They also published large datasets (≈ 96 million frames) built using ViPE to accelerate training of spatial AI systems.
[Embedding Models] IBM has released two new embedding models, granite-embedding-english-r2 (149M) and granite-embedding-small-english-r2 (47M), built on ModernBERT with support for 8192-token context, optimized attention mechanisms, and FlashAttention 2. Both models deliver strong performance on benchmarks like MTEB, BEIR, CoIR, and MLDR, while maintaining high throughput on GPUs and CPUs, making them ideal for large-scale retrieval and RAG pipelines. Crucially, they are released under the Apache 2.0 license, ensuring unrestricted commercial use.
[Inference] BentoML Released llm-optimizer: An Open-Source AI Tool for Benchmarking and Optimizing LLM Inference. It automates configuration testing across frameworks like vLLM and SGLang, applies constraints such as latency or throughput targets, and delivers reproducible results through interactive dashboards. Alongside, the LLM Performance Explorer offers pre-computed benchmarks for popular models, enabling easier comparison and analysis. Together, they reduce trial-and-error in LLM optimization and bring transparency and consistency to performance evaluation
[FREE WEBINAR] Contain Lateral Movement and Protect Your Clients using NetBird Integration with Acronis (Sponsored)
[Voice AI & Open Source] UT Austin and ServiceNow Research Team Releases AU-Harness: An Open-Source Toolkit for Holistic Evaluation of Audio LLMs. It delivers up to 127% faster throughput and 59% lower latency through vLLM integration. It standardizes evaluation with configurable prompts and metrics, supports multi-turn dialogue, and spans six task categories—covering 50+ datasets, 380+ subsets, 21 tasks, and 9 metrics. Uniquely, it introduces LLM-Adaptive Diarization and Spoken Language Reasoning tasks, exposing gaps in temporal understanding and complex spoken instruction following.
[Open Model] Meta AI Released MobileLLM-R1: A Edge Reasoning Model with less than 1B Parameters and Achieves 2x–5x Performance Boost Over Other Fully Open-Source AI Models. The flagship 950M model was trained on fewer than 5T tokens—about 1/9 the data of Qwen3-0.6B—yet matches or surpasses it on reasoning benchmarks (74.0 vs 73.0 on MATH500) and delivers 2×–5× gains over SmolLM2-1.7B and OLMo-1B in math accuracy. With optimizations like grouped-query attention and block-wise weight sharing, MobileLLM-R1 demonstrates that compact, domain-specialized LLMs can achieve state-of-the-art reasoning performance while remaining efficient for edge deployment.
Editor’s Pick
[Private AI] Google AI Releases VaultGemma: The Largest and Most Capable Open Model (1B-parameters) Trained from Scratch with Differential Privacy. Built on the Gemma architecture with 26 transformer layers and a 1024-token context, it was trained on 13T filtered tokens using DP-SGD and a TPUv6e cluster of 2048 chips. The model provides a strong privacy guarantee of (ε ≤ 2.0, δ ≤ 1.1e−10) and shows no detectable training data leakage. While its benchmark scores (ARC-C 26.45, PIQA 68.0, TriviaQA 11.24) trail non-private counterparts, performance is on par with older GPT-2-scale models, marking a critical milestone in scaling privacy-preserving AI.
From our Sponsor
[FREE WEBINAR] Contain Lateral Movement and Protect Your Clients using NetBird Integration with Acronis.
Date and Time: 30th September, 5:00 PM CET [45 minutes with Q&A]
Adversaries are increasingly targeting Managed Service Providers (MSPs) with sophisticated tactics and techniques. According to the Acronis Cyberthreats Report, H2 2024, sophisticated APT-linked ransomware groups are eyeing MSPs—exploiting PowerShells, weak RDP passwords, unpatched devices, and compromised VPN credentials. The adversaries are relentless. But how can MSPs shift from a reactive approach and get proactive to reduce the blast radius?
Join us for an exclusive session with James Abercrombie, Technology Evangelist, Acronis, and Naren Vaideeswaran, Head of Product Marketing, NetBird, as they discuss how the integration works, the benefits, and how MSPs can effectively shrink the attack surface.
In this webinar, you will learn:
The impact of lateral movement and how ransomware is affecting businesses and reputation
How a multi-layered defense paves the way for effective prevention, detection, and disaster recovery readiness
How NetBird and Acronis integrate to contain evolving threats and protect your business.
(Sponsored)
AI Agents, MCP & Voice AI
[Agentic AI] OpenAI Releases GPT-5-Codex — a version of GPT-5 further optimized for agentic coding in Codex
[AI Productivity] Sanebox is a super smart AI Tool that automatically filters unimportant emails, letting you focus on what matters. (Sponsored)