Microsoft’s model-routing strategy is getting more explicit: Satya Nadella said MAI models are already taking traffic in GitHub Copilot, Excel, and Outlook, with Copilot Chat and PowerPoint next once the first-party models clear the bar. In the same stretch, OpenAI launched GPT-5.6 GA with ultra multi-agent mode, and Microsoft 365 Copilot made GPT-5.6 the preferred model for Word, Excel, PowerPoint, Chat, and Cowork. Sources: Satya on X ; OpenAI GPT-5.6 ; Microsoft 365 Copilot GPT-5.6
Anthropic Claude Opus 5 is now available across Microsoft 365 Copilot, Microsoft Foundry, and GitHub Copilot. Microsoft frames it as better fit for complex, long-running, tool-heavy work, and the GitHub Copilot / Reddit chatter shows the usual immediate “which model should I actually use?” reaction. Sources: M365 Copilot ; Foundry ; GitHub changelog ; Reddit ; HN
Microsoft Agent Framework 1.12.0 is a real platform release, not just a version bump: app-owned A2A/MCP hosting helpers, Responses protocol helpers, Telegram hosting samples, a Microsoft OTel observability sample, and stable graduation for harness agent, ToolApprovalMiddleware, FileMemoryProvider, and declarative workflows. Source: GitHub releases
Google DeepMind shipped Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber with a clear agentic-scale pitch: lower latency, fewer tool calls, built-in computer use, and a cyber-focused variant for defenders. HN is already debating where it sits for programming and long-horizon tool use. Sources: Google DeepMind ; HN
OpenAI Presence launched as a deployed enterprise product for voice/chat agents that can answer questions, use company systems, take approved actions, and escalate to humans. This reads more like an actual agent operations product than a marketing demo. Sources: OpenAI Presence ; HN