Universal-3.6 Pro tops Coval's 29-model streaming STT benchmark with lowest WER and ~106ms latency
AssemblyAI · x · 2026-10-17
Coval's latest public streaming Speech-to-Text benchmark (Oct 2026, 29 models) ranks AssemblyAI's Universal-3.6 Pro with the lowest word error rate (2%) while keeping a 106ms P50 time-to-final-transcript—placing it alone in the optimal lower-left corner. Its predecessor Universal-3.5 Pro sits at 2.5% WER / 108ms.
Median time to final transcript dropped about a third this week for both: 3.6 Pro went from 153ms to 106ms, 3.5 Pro from 156ms to 108ms. Compared models include Gemini 3.5 Transcribe Live, GPT-4o Transcribe, Chirp 3, Whisper Large v3, Nova 3, and Qwen3 ASR. AssemblyAI credits Coval for running these benchmarks in public every week.
More from Models
- Dev ditches Gemini 4 Argon for Grok 4.5, which eats 96.6% of his Cursor tokens — gaganghotra_ · 2026-10-17
- LMArena opens Code Arena WebDev for blind head-to-head model coding battles — arena · 2026-10-17
- Unverified: GPT-6 Pro said to launch parallel research subagents, 50-100 weekly messages — TheMoonMidas · 2026-10-17
- Gemini 4 Argon costs 43¢ vs Opus 5.5's $2.43 on same build prompt — MattVidPro · 2026-10-17
- Perplexity quietly retires GLM 5.3, user still used it two minutes earlier — ChrisUniverse · 2026-10-17
- NVIDIA Launches Fastino Training API: Pay Per Token for SFT and RL — NVIDIAAI · 2026-10-17