AI Weekly: Voice Models, Custom Silicon, MCP Goes Enterprise (May 7–13, 2026)

Short summary

OpenAI shipped three voice models and GPT-5.5; Anthropic doubled Claude Code rate limits leveraging SpaceX's 300MW Colossus compute. Cursor expanded to Microsoft Teams with agent delegation and launched a security-review beta; GitHub Copilot transitioned to usage-based billing with a 15x→27x Opus multiplier jump. The underlying pattern: AI coding tools, custom silicon infrastructure, and enterprise standards (MCP) are maturing in parallel, each reshaping competitive positioning and pricing strategy.

• •OpenAI added three voice models and GPT-5.5 security variant; Anthropic doubled Claude Code rate limits using 300MW SpaceX Colossus compute
• •Cursor available in Microsoft Teams with @cursor agent delegation; launched Security Review beta with PR vulnerability scanning and remediation guidance
• •GitHub Copilot moved to usage-based billing with 27x Opus multiplier; Claude Code now runs in five surfaces (shell, VS Code, JetBrains, GitHub Actions, web/mobile)

Generated with AI, which can make mistakes.