Moonshot AI’s Kimi K3 is rewriting open-source benchmarks, taking the top spot among open-weight models across major leaderboards. With its massive 2.8-trillion-parameter MoE architecture and native vision capabilities, Kimi K3 is closing the gap with proprietary frontier models. Meanwhile, Microsoft is bolstering enterprise defense with MAI-Cyber-1-Flash—its specialized 137B parameter in-house security model designed to power its MDASH multi-agent vulnerability orchestration engine.

We’ll talk about:

• Ranking #1 among open-weight models in Agent Arena and leading the Frontend Code Arena with 1,682 points.
• A specialized 137B parameter model integrated into MDASH, driving CyberGym benchmark scores up to 95.95% while halving operational costs.
• OpenAI enabling persistent auth across sessions to ...