Higgs Realtime Duplex takes in audio, video and text continuously, and speaks, thinks, calls tools and renders a live face while it is still listening.

In Hume AI's independent evaluation of eleven voice-cloning models, Higgs TTS 3 placed fourth on speaker similarity and scored highest on accented reference voices.

A benchmark for post-interruption recovery in voice agents — whether what the model says after you cut in resumes the workflow correctly. 428 interruption samples, 6 types, 10 enterprise domains, closed- and open-weight models.

A benchmark for conversational proactivity in LLMs — noticing and acting on what the user implied but never said. 198 curated dialogues, 624 trigger points, 16 models, and a leaderboard where Recovery proves dramatically hard.

Since the release of Higgs Llama 2, we have received much positive feedback from the community. While we are amazed by the community's creativity in utilizing our model, we realize the importance of providing an automated benchmark to effectively evaluate large language model (LLM)'s roleplaying capability.

At Boson AI, we are working on intelligent agents that can serve as human companions and helpers. Today we are excited to share Higgs Llama 2 70B, a new model that significantly improves upon its predecessor. It narrows the gap to the very best proprietary models on benchmarks relevant for dialog interaction and understanding.