March 3 AI Briefing · Issue 78
The Qwen 3.5 series of compact models (0.8B–9B) has seen widespread deployment, supporting multi-platform inference on MLX, Ollama, and LM Studio—and even running natively on edge devices like the iPhone 17 and routers. Meanwhile, Claude Code launched a free voice mode, and OpenClaw...
Editorial standards and source policy: Editorial standards, Team. Content links to primary sources; see Methodology.
## 🔍 Key Insights
The **Qwen 3.5 series of compact models** (0.8B–9B) are rapidly entering real-world deployment, supporting multi-platform inference on **MLX, Ollama, and LM Studio**, enabling native execution on **edge devices such as the iPhone 17 and home routers**. Meanwhile, **Claude Code** has launched a free **voice mode**, and **OpenClaw** has accelerated its iteration cycle while expanding globally to **80+ in-person Meetups**—marking AI Agents’ decisive shift from “parameter races” toward **physical-world integration** and **scenario-specific delivery**.
## 🚀 Major Updates
- **Qwen 3.5 Full Series Now Available on Ollama & MLX**: Four models—from 0.8B to 9B—officially support tool calling and multimodal capabilities; native execution on Apple Silicon is now seamless.
- **Claude Code Launches Free Voice Mode**: Developers can now issue voice commands to directly interact with and manipulate codebases—CLI integration requires zero additional cost.
- **Xiaomi’s MiMo Large Model Debuts at MWC**: Powers Xiaomi’s “human–vehicle–home” integrated ecosystem, signaling a strategic pivot in AI competition—from chat interfaces to **real-world system control**.
- **Shannon Open-Sources an AI Hacker Agent**: A fully autonomous web vulnerability scanner and exploit engine, achieving **zero false positives** and top-tier benchmark accuracy.
- **OpenClaw v2026.3.2 Released**: Introduces Telegram-based real-time streaming, native PDF processing, and major security enhancements.
- **“Cai Dazi” Launches China-First A-Share AI Agent**: Evolves from a Q&A assistant into a collaborative multi-agent orchestration platform—focused on **companion-style empowerment** across the financial decision-making chain.
- **Liquid AI Unveils LFM2.5-1.2B-Thinking**: An edge-optimized reasoning model with 1.17B parameters, engineered for ultra-low-memory environments and high-speed local inference.
- **Meituan’s AI Browser Tabbit Officially Launched**: Replaces traditional task flows with **scenario-driven automation**, demonstrating strong practical utility in office productivity.
The Qwen 3.5 series of compact models (0.8B–9B) are rapidly entering real-world deployment, supporting multi-platform inference on MLX, Ollama, and LM Studio, enabling native execution on edge devices such as the iPhone 17 and home routers. Meanwhile, Claude Code has launched a free voice mode, and OpenClaw has accelerated its iteration cycle while expanding globally to 80+ in-person Meetups—marking AI Agents’ decisive shift from “parameter races” toward physical-world integration and scenario-specific delivery.
🚀 Major Updates
- Qwen 3.5 Full Series Now Available on Ollama & MLX: Four models—from 0.8B to 9B—officially support tool calling and multimodal capabilities; native execution on Apple Silicon is now seamless.
- Claude Code Launches Free Voice Mode: Developers can now issue voice commands to directly interact with and manipulate codebases—CLI integration requires zero additional cost.
- Xiaomi’s MiMo Large Model Debuts at MWC: Powers Xiaomi’s “human–vehicle–home” integrated ecosystem, signaling a strategic pivot in AI competition—from chat interfaces to real-world system control.
- Shannon Open-Sources an AI Hacker Agent: A fully autonomous web vulnerability scanner and exploit engine, achieving zero false positives and top-tier benchmark accuracy.
- OpenClaw v2026.3.2 Released: Introduces Telegram-based real-time streaming, native PDF processing, and major security enhancements.
- “Cai Dazi” Launches China-First A-Share AI Agent: Evolves from a Q&A assistant into a collaborative multi-agent orchestration platform—focused on companion-style empowerment across the financial decision-making chain.
- Liquid AI Unveils LFM2.5-1.2B-Thinking: An edge-optimized reasoning model with 1.17B parameters, engineered for ultra-low-memory environments and high-speed local inference.
- Meituan’s AI Browser Tabbit Officially Launched: Replaces traditional task flows with scenario-driven automation, demonstrating strong practical utility in office productivity.