Anthropic has officially launched Claude 4.5 "Global Agent", its most advanced agentic AI system to date. Powered by multi-agent orchestration, it supports collaborative task execution via Dev Team mode and persistent agent threads. Built on Claude Sonnet 4.5, it scores 76% on conversational analytics benchmarks and offers built-in VM isolation and safety features, enabling autonomous task completion for up to 30 hours.

Alibaba has officially launched the full public beta of Qwen 3.5, its latest flagship multimodal AI model. Featuring ultra-long video perception supporting up to two hours of continuous footage, it adopts a 397B-parameter sparse MoE architecture with a 1M-token context window and multilingual OCR. Delivering 19x faster inference and 60% lower cost, Qwen 3.5 serves as a strong competitor to GPT-5.2 and Claude 4.5, available via Qwen Chat, Alibaba Cloud API, and open-weight downloads.

OpenClaw, the world's largest open-source AI agent framework (over 214,000 GitHub stars), is facing a critical security crisis: over 230,000 AI instances are exposed on the public internet with unencrypted communications. China's Ministry of State Security issued an urgent warning, noting the framework’s fundamental vulnerability enables remote code execution via prompt injection, classified as "critical". Global organizations are urged to audit their deployments and implement security measures or adopt alternatives like NemoClaw.

Microsoft Research and Huawei jointly announced a key commercial milestone for 1-bit large language models (LLMs), stating that BitNet b1.58 technology has realized the efficient deployment of large-scale models on edge devices equipped with Huawei's latest Kirin chips and Qualcomm's Snapdragon platforms. Supported by BitNet's innovative ternary weight architecture {-1, 0, 1}, the technology reduces the model's memory footprint by over 70% and consumes only 0.028 joules per inference, enabling large-scale LLMs to run on consumer mobile hardware without relying on cloud connectivity. This breakthrough optimizes the economics of on-device AI, making high-performance model capabilities accessible on smartphones and tablets without the need for data center infrastructure or network latency.

**Together AI**, as a diamond sponsor of **NVIDIA GTC 2026**, is showcasing its latest research and product innovations at Booth #1213 in San Jose from March 16 to 19. Today’s updates focus on open-source LLMs, voice AI capabilities, production-grade inference, and AI factory infrastructure. Key announcements include **FlashAttention-4** (up to 1.3× faster than cuDNN on NVIDIA Blackwell), the open-source **ThunderAgent** for agentic workloads (delivering a 3.6× throughput improvement), the **ATLAS-2** adaptive learning speculator, and a full-featured voice AI stack supporting real-time speech-to-text and text-to-speech. Together AI demonstrates how enterprises can transition from AI experiments to production deployment in minutes using its GPU clusters and inference platform.

NVIDIA officially unveiled NemoClaw at GTC 2026 on March 16, an enhanced open-source AI agent stack designed to bring enterprise-grade security, privacy, and multi-agent orchestration to the OpenClaw ecosystem. NemoClaw installs in a single command and combines the OpenClaw agent platform with NVIDIA's Agent Toolkit components, including OpenShell for isolated sandbox execution and AI-Q for building reasoning agents. The stack addresses the critical security concerns that have held back enterprise adoption of autonomous AI agents, offering policy-based guardrails, data privacy controls, and support for NVIDIA's Nemotron open models across deployment environments from RTX PCs to DGX Spark and DGX Station.

Meta Platforms has delayed the release of its next-generation AI model codenamed "Avocado" from March to at least May 2026, after internal testing revealed the model's performance falls short of leading competitors including OpenAI's GPT, Anthropic's Claude, and Google's latest Gemini models. The New York Times first reported that Meta's AI division leaders are now considering temporarily licensing Google's Gemini to power Meta's AI products. The delay comes as Meta plans to invest up to $135 billion in AI infrastructure in 2026 and reportedly considers layoffs of 20% or more of its workforce to fund aggressive AI spending.

The 2026 AI funding landscape has shattered all records, with OpenAI leading the pack at $174B+ total funding following its massive $110B February round. xAI secured $20B in January's upsized Series E, while Anthropic reached $69.2B total after its $30B Series G at $380B valuation. February 2026 became the biggest month in venture capital history with $189B global funding, of which $171B went to AI startups. Figure AI ($3.2B+) and Perplexity ($2.6B+) round out the top five, as AI startups captured over 90% of venture funding in the record-breaking month.

Zhipu AI (02513.HK) officially released GLM-5-Turbo on March 16, 2026 — the world's first foundation model specifically optimized for OpenClaw ("Lobster") scenarios. The model is enhanced from the training stage for core Agent capabilities including tool calling, instruction following, scheduled tasks, and long-chain execution. Simultaneously, Zhipu raised GLM-5-Turbo API prices by 20% and launched "Lobster Plan" subscriptions starting at ¥39 for 35 million tokens. GLM-5-Turbo has been integrated into iSoftStone's Mechrevo "Lobster Box," creating a native AI Agent terminal experience. Zhipu shares surged up to 16% intraday following the announcement.

OpenAI has released GPT-5.4 — simultaneously its most capable frontier model and its most decisive statement yet that AI has crossed from "answering questions" into "doing the work." Scoring 83% on GDPval's 44-occupation knowledge work benchmark, achieving 75% on OSWorld computer-control tasks (surpassing the human benchmark of 72.4%), slashing hallucinations by 33%, introducing native computer use for the first time in a general-purpose model, and launching Tool Search to cut token costs 47% in large agent ecosystems — GPT-5.4 is not an incremental update. It is OpenAI's declaration that professional workflows now have a new default executor.

NVIDIA GTC 2026 — the world's most closely watched AI conference — has officially opened in San Jose, California, with founder and CEO Jensen Huang delivering a landmark keynote from the SAP Center to 30,000 attendees from 190 countries. Huang declared that AI is no longer an application or a model: "It is essential infrastructure. Every company will use it. Every nation will build it." The conference formally unveils the Vera Rubin platform now in full production, the anticipated NemoClaw open-source enterprise AI agent platform, the $20 billion Groq LPU inference integration, Nemotron 3 Super for agentic reasoning, physical AI leadership across robotics and autonomous systems, and a five-layer AI technology stack that Huang says is powering "one of the largest infrastructure expansions in history."

Microsoft has unveiled Copilot Cowork — the most significant evolution of Microsoft 365 Copilot since its launch — a fully autonomous, multi-step task execution agent powered by Anthropic's Claude model and the new Work IQ intelligence layer. Currently in Research Preview with select enterprise customers and expanding via the Frontier program in late March 2026, Cowork marks Microsoft's definitive shift from "AI that answers" to "AI that executes" — running long-horizon tasks across Outlook, Teams, Word, Excel, and PowerPoint with full enterprise security, human checkpoints, and auditable action trails.