OpenAI released gpt-realtime, a speech-to-speech model with improved instruction-following capabilities and more natural-sounding voices. The model integrates with the Realtime API, which now supports MCP, image inputs, and SIP for phone connectivity. Facing competition from ElevenLabs and others, OpenAI reduced prices while enhancing enterprise features for customer service applications.

Stanford research analyzing ADP payroll data reveals young workers in AI-exposed fields have experienced a 13% employment decline since 2022, while older workers remain stable. Jobs requiring experience show resilience against AI disruption, whereas roles dependent on formal education face greater impact. This trend explains stagnant employment growth among young workers despite overall market resilience.

No-code website builder Framer has reached a $2 billion valuation after raising $100 million in Series D funding. The company reports 500,000 monthly active users and $50 million in annual recurring revenue. Framer is investing in enterprise growth and AI capabilities, with business customers now comprising the majority of new users.

Nous Research released Hermes 4, open-source AI models matching proprietary systems while offering minimal content restrictions. The models feature hybrid reasoning capabilities, outperforming ChatGPT and Claude on math benchmarks and RefusalBench. Trained with novel DataForge and Atropos systems, Hermes 4 represents a philosophical shift toward user control versus corporate content moderation.

Microsoft announced its first in-house AI models: MAI-Voice-1 and MAI-1-preview. MAI-Voice-1 generates audio rapidly on a single GPU, powering features like Copilot Daily. MAI-1-preview, trained on 15,000 Nvidia GPUs, handles instructions and everyday queries. Microsoft plans to integrate these models into Copilot, which currently relies on OpenAI technology.