AssemblyAI launches Universal-3.6 Pro Realtime, tops two independent streaming STT benchmarks
AssemblyAI · x · 2026-09-30
AssemblyAI released Universal-3.6 Pro Realtime, a streaming speech model aimed at voice agents. It leads two independent benchmarks: it sits on the Pareto frontier of @trydaily's Piecat open STT benchmark and has the lowest word error rate of any streaming model on @covaldev's live leaderboard, tested on real voice-agent audio. The model supports 99 languages and is available via the company's API platform.
More from Models
- Oxmiq Open-Sources NVFP4 Quantized Hy4 Model, Runs on RTX 6000 Pro — RajaXg · 2026-09-30
- Cost Pareto-dominant, but second only to Astra on output-token efficiency — adonis_singh · 2026-09-30
- GPT-6.1-Sol Takes #2 on eyebench-v3, ~8x Cheaper Than Opus-5.5 — adonis_singh · 2026-09-30
- GPT-6.1-Sol Takes #2 on eyebench-v3, ~8x Cheaper Than Opus-5.5 — adonis_singh · 2026-09-30
- Writing benchmark: GPT-6.1 Sol regresses 153 Elo below GPT-6 Sol while costing more — OnlyProggingForFun · 2026-09-30
- Ethan Mollick: OpenAI has launched and semi-abandoned a dev ecosystem every year since 2023 — emollick · 2026-09-30