AssemblyAI Universal-3 Pro (released Feb 2026) is the industry's first promptable speech language model for Voice AI, delivering context-aware, high-accuracy transcription with natural language instructions. Prompt for disfluencies, keyterms (up to 1,000), speaker roles, audio events ([beep], laughter), verbatim/clean modes, code-switching (English/Spanish), and domain-specific accuracy (medical/legal/business). Achieves 94.07% word accuracy on real-world audio—topping competitors like ElevenLabs/OpenAI— at 35-50% lower cost ($0.21/hr base + add-ons). Free all February; ideal for call centers, healthcare, legal, podcasts needing precise, customizable speech-to-text.