🤖 AI Agents Weekly: Claude Agent SDK, Sora 2, Claude Sonnet 4.5, Microsoft Agent Framework, GLM-4.6, Agentic Commerce Protocol
Claude Agent SDK, Sora 2, Claude Sonnet 4.5, Microsoft Agent Framework, GLM-4.6, Agentic Commerce Protocol
In today’s issue:
Anthropic introduced Claude Sonnet 4.5
OpenAI announced Sora 2
Microsoft announced the Microsoft Agent Framework
Thinking Machines announced Tinker
New research on training agents inside scalable world models
Zhipu AI has released GLM-4.6
OpenAI’s Agentic Commerce Protocol (ACP)
Anthropic announces Claude Agent SDK
…and much more.
Top Stories
Claude 4.5
Anthropic introduced Claude Sonnet 4.5, positioning it as the strongest coding and agent-building model to date, with major advances in reasoning, math, and computer use. It arrives alongside new developer tools, upgrades to Claude Code, and the launch of the Claude Agent SDK.
Key highlights:
Frontier performance – Claude Sonnet 4.5 leads on SWE-bench Verified (real-world coding tasks) with up to 82% accuracy under high-compute settings, and on OSWorld (computer-use benchmark) it jumped from 42.2% → 61.4% in four months. It maintains task focus for 30+ hours.
Product upgrades – Claude Code gains checkpoints, refreshed terminal UI, VS Code extension, and memory/context editing via API. Claude apps now support code execution and file creation (spreadsheets, slides, docs). A Chrome extension ships to Max users.
Claude Agent SDK – Anthropic is releasing the same infrastructure behind Claude Code to all developers, enabling custom agent design with built-in memory, permissions, and subagent coordination.
Domain strength – Experts report large gains in finance, law, medicine, and STEM compared to Opus 4.1. Early customers confirm stronger domain-specific reasoning.
Research preview – “Imagine with Claude” demo shows real-time adaptive software generation. Available for Max subscribers for five days.
Pricing and availability – Claude Sonnet 4.5 is available globally at the same cost as Sonnet 4 ($3/$15 per million tokens) and is a drop-in API replacement.
Keep reading with a 7-day free trial
Subscribe to AI Newsletter to keep reading this post and get 7 days of free access to the full post archives.