AssemblyAI is introducing two major product updates that are set to redefine industry standards for Speech AI capabilities. The first product announcement is the industry's first production-ready speech language model, called Slam-1, which enables rapidly customizable speech-to-text and speech understanding tasks via prompting. Slam-1 optimizes accuracy for specific applications and industries by leveraging a pretrained LLM and fine-tuning an adapter layer to bridge the gap between the high-performance speech encoder and the language model's semantic space. The model has shown promising results in reducing miss-detection rates of specified named entities by 74% and medical term miss-detection rate by 59%. Slam-1 is currently in private beta with select partners and will be launched in the coming weeks. Additionally, AssemblyAI is announcing its new Streaming API for Voice Agents use cases, featuring low latency, intelligent endpointing, high accuracy, and robustness to background speakers/noise in real-world environments. The new Streaming API is set to empower customers to build voice agents that don't just respond quickly, but actually understand what people are saying.