Pine AI Posts 75.4% State of the Art on the τ³-Voice Leaderboard
τ-Bench Leaderboard / Hacker News·low signal
Pine AI reached 75.4% on the τ³-Voice leaderboard, a new state of the art for the voice-agent benchmark, surfaced via Hacker News. τ³-Voice extends the τ-bench tool-use evaluation family into the voice modality, where agents have to hold state across a spoken conversation while calling tools. Voice agent benchmarks are still thin enough that a single leaderboard result moves the field's reference point.