AI Dev and Latest Releases
[Deep Research Agents] NVIDIA AI Releases Universal Deep Research (UDR): A Prototype Framework for Scalable and Auditable Deep Research Agents. Unlike existing deep research tools that enforce rigid, model-tied workflows, UDR decouples strategy from model, allowing users to design, edit, and execute domain-specific research strategies without retraining. By converting natural language strategies into executable code, orchestrating workflows at the system level, and using LLMs only for localized reasoning, UDR enables flexible, auditable, and efficient research automation across domains such as scientific discovery, business intelligence, and technical due diligence
[Reasoning Open Model] Baidu has released ERNIE-4.5-21B-A3B-Thinking, a reasoning-optimized Mixture-of-Experts model with 21B parameters (3B active per token), supporting 128K context length for long-document reasoning and multi-step workflows. It integrates tool and function calling, excels in mathematics, science, logic, and coding benchmarks, and can be deployed on a single 80GB GPU with quantization for efficiency. The model supports English and Chinese, is released under the Apache-2.0 license, and is available on Hugging Face, positioning it as a commercial-friendly, long-context reasoning model that balances performance with deployment practicality.
[FREE WEBINAR] Contain Lateral Movement and Protect Your Clients using NetBird Integration with Acronis
[ASR Voice AI] Alibaba Qwen Team Releases Qwen3-ASR: A New Speech Recognition Model Built Upon Qwen3-Omni Achieving Robust Speech Recogition Performance. Qwen3-ASR supports 11 languages with automatic detection, context-aware transcription, and robust performance in noisy, low-quality, and far-field audio. It achieves under 8% word error rate even on songs and raps, and allows custom vocabulary injection for domain-specific terms, making it suitable for applications in education, media, and customer service.
[Voice AI] Deepdub Introduces Lightning 2.5: A Real-Time AI Voice Model With 2.8x Throughput Gains for Scalable AI Agents and Enterprise AI. The system delivers 2.8× higher throughput and 5× efficiency gains over previous iterations, optimized for NVIDIA GPU-accelerated infrastructure to reduce latency and cost per output. Designed to preserve natural prosody, emotional nuance, and multilingual fidelity, Lightning 2.5 positions Deepdub as a competitive player in low-latency speech synthesis, though detailed benchmarks on latency, architecture, and language coverage remain undisclosed.
[Voice AI] TwinMind Introduces Ear-3 Model: A New Voice AI Model that Sets New Industry Records in Accuracy, Speaker Labeling, Languages and Price. It sets industry records with 94.74% accuracy (5.26% WER), 3.8% diarization error rate, support for 140+ languages, and a low cost of $0.23/hour. Built from a blend of open-source models and curated training data, Ear-3 is positioned against services from Deepgram, AssemblyAI, Speechmatics, OpenAI, and others. While offering strong gains in accuracy, language coverage, and pricing, the model requires cloud deployment, raising questions about privacy, offline usability, and real-world robustness across diverse environments.
Editor’s Pick
[Open Source] MBZUAI Researchers Release K2 Think: A 32B Open-Source System for Advanced AI Reasoning and Outperforms 20x Larger Reasoning Models. It integrates chain-of-thought supervised fine-tuning, reinforcement learning with verifiable rewards, agentic planning, and wafer-scale inference optimizations. It achieves frontier-level performance—scoring 90.83 on AIME’24, 81.24 on AIME’25, and 63.97 on LiveCodeBench—while reducing token usage by up to 11.7% and delivering ~2,000 tokens per second on Cerebras hardware. Released with full transparency, including weights, training data, and code, K2 Think shows how carefully engineered mid-scale models can rival much larger proprietary systems in reasoning efficiency and accuracy.
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
AI Agents and MCP
[AI Productivity] Sanebox* is a super smart AI Tool that automatically filters unimportant emails, letting you focus on what matters.