Hume EVI 3

EVI 3 (pronounced 'Evee Three') is Hume AI's third-generation Empathic Voice Interface, launched May 2025. It's a speech-to-speech conversational model built around emotional intelligence: it reads the user's tone and modulates its own responses with natural pauses, hesitations, and expressive delivery. Its standout feature is natural-language voice creation — describe a personality ('warm confident storyteller with a French accent') and it generates a matching custom voice in seconds, no voice-design expertise required. Hume's internal benchmarks showed it beating GPT-4o and Gemini Live on naturalness, expressiveness, and emotion understanding, with ~300ms response latency.

The honest picture: those benchmark wins are Hume's own numbers, not independent evaluations, and one hands-on review found the voice-cloning output more 'audio cartoon' than uncanny. At launch it was English-focused with other languages promised later. API access for developers rolled out in the weeks after launch, and pricing was still being finalized — this is a startup-scale product, not a hyperscaler platform, so enterprise buyers should validate SLAs and uptime before committing.