The AI Wire

azure ▸ product· filed 22 Aug 2026

Microsoft Foundry Adds DeepSeek-V4-Flash-0731 and NVIDIA Nemotron 3.5 Lightning Models

Microsoft Foundry is adding DeepSeek-V4-Flash-0731 and NVIDIA Nemotron 3.5 Lightning to its catalog, both optimized for agentic workloads. The models are available through multiple deployment paths, including Azure, Fireworks, and Hugging Face, with varying formats and pricing. Microsoft reports significant benchmark improvements for DeepSeek, though these figures are not independently verified.

azure ▸ product· filed 22 Aug 2026

Microsoft expands Foundry model router to additional regions and refreshes model pool

Microsoft expanded its Foundry model router to more regions and updated its supported model list, adding Claude Opus 4.8 and GPT-5.6 while removing four older models. The router's endpoint remains stable through these changes, and Microsoft describes usage patterns including A/B testing and continuous routing. The company recommends monitoring routing traces and outcomes to confirm performance targets.

youtube.com ▸ product· filed 21 Aug 2026

Microsoft's New Azure AI Exam Covers Foundry's Model Catalog, Router, and Agents

This survey of Azure AI Foundry maps its core components to the new exam blueprint, covering the model catalog, router, and agent framework. It also explores deployment options, API differences, and cost considerations, with examples from real usage.

azure ▸ product· filed 17 Aug 2026

Anthropic extends five agent-building capabilities to Azure-hosted Claude deployments in Microsoft Foundry

Anthropic has extended five agent-building capabilities—including structured outputs, web search, and an MCP connector—to Claude models on Azure-hosted Foundry deployments, eliminating the need for workarounds in residency-constrained environments. The features address common LLM pipeline issues like malformed JSON, token inefficiency, and tool-selection degradation. Pricing and data-handling details are included, with recommendations to review certain features for regulated workloads.

learn.microsoft.com ▸ product· filed 17 Aug 2026

Microsoft Foundry Agent Service provides managed runtime for containerized AI agents

Microsoft Foundry Agent Service offers a managed runtime for containerized AI agents on Azure, handling deployment, scaling, and session persistence. It supports multiple agent frameworks and protocols, with per-session sandboxing and immutable versioning. Users retain responsibility for agent behavior and compliance.

blogs.nvidia.com ▸ models· filed 17 Aug 2026

NVIDIA adds Nemotron 3.5 Lightning and NeMo Switchyard to its open-model lineup

NVIDIA has released Nemotron 3.5 Lightning, a 30-billion-parameter mixture-of-experts model for high-volume agentic AI workloads, and NeMo Switchyard, an open-source library for routing requests across models. The company reports faster output and lower cost based on its own benchmarks and partner-reported results; one partner's routing test also showed an accuracy tradeoff.

theguardian.com ▸ industry· filed 17 Aug 2026

Zuckerberg Essay Defends Open-Weight AI, Opposes Regulation

Mark Zuckerberg published an essay advocating open-weight AI models and opposing regulation, positioning Meta as a democratic alternative to closed systems from rivals. He offered few specifics on safety testing, datacenter policy, or labor impacts, while downplaying concerns about surveillance and weapons. The essay coincided with the release of a new Meta model, Muse Glimmer.

anthropic ▸ policy· filed 15 Aug 2026

Anthropic to Add Statistical Watermark to Claude Text

Anthropic will embed a statistical watermark in Claude-generated text, citing the EU AI Act as the driver. The method preserves output quality and readability, but has limitations with short, factual, or code-heavy passages. A detection API is planned, though access details remain undecided.

huggingface ▸ industry· filed 14 Aug 2026

Hugging Face Hub analysis: Chinese labs dominate largest open models as small models drive most downloads

Hugging Face's mid-year analysis of Hub activity from January through August 2026 finds sharp divides between attention and adoption: 1.5% of repositories account for 99.2% of downloads, and no 2026 model reaches the download top 25. Chinese labs released the largest open-weight models nearly every month, while Qwen became the most common base for derivatives and models under 1 billion parameters took 83% of all-time downloads. The report also notes coding agents are emerging as a direct source of Hub traffic.

z.ai ▸ product· filed 14 Aug 2026

Z.ai's GLM-5.3: open-weights model with post-training gains in agentic coding and vulnerability exploitation

GLM-5.3, Z.ai's open-weights model built on the GLM-5.2 base, achieves post-training improvements in agentic coding and vulnerability exploitation, showing a 50% gain on Z.ai's Code Bench and identifying over 2,400 vulnerabilities. However, exploitation scores remain below closed models, and weights are due in two weeks after safety checks.