HeyGen Voice tops Artificial Analysis TTS Arena with Elo 1201, beats Qwen and ElevenLabs
ArtificialAnlys · x · 2026-10-10
Artificial Analysis updated its Controlled Voice TTS Arena: HeyGen's first text-to-speech model, HeyGen Voice, takes the #1 spot with an Elo of 1201 across 1,468 arena appearances, ahead of Alibaba's Qwen-Audio-3.1-TTS-Plus (1182), ElevenLabs' Eleven v4 Turbo (1166) and Eleven v4 (1162).
- Evaluated on the same 8 cloned voices across US/UK English; ranks #1 in Assistants/Customer Service and English (UK), #2 in Knowledge Sharing
- Pronunciation Robustness: 83.1%, #10 of 29 models; Preserving Exact Sequences 83.3%, #2 behind SpaceXAI TTS
- Pricing: $30/1M characters — below Eleven v4 Turbo ($40) and Eleven v4 ($80), above Qwen ($19.3)
- Speed: 40 characters per second of generation time vs 76 for Eleven v4 and 46 for Gemini 3.8 Flash TTS
Related event: HeyGen Voice Tops Blind-Test TTS Leaderboard, API at Half Price(7 posts)→
More from Models
- Integer multiplication tracker for LLMs adds zoom to visualize recent progress — RexDouglass · 2026-10-10
- Cloudflare ships open-weight Clef-omni with audio/video input, cuts Clef-flash to $0.038/M tokens — Cloudflare Blog · 2026-10-10
- Radiologist pressures AI three times to sign off a benign breast report — it holds firm — FellMentKE · 2026-10-10
- Radiologist tests medical AI: model refuses wrong BI-RADS and demands more evidence — FellMentKE · 2026-10-10
- OpenAI pushes 722 math manuscripts to GitHub from a model nobody can use — thursdai_pod · 2026-10-10
- Text-to-LoRA works: predicting LoRA distributions to scale inference by sampling weights — akyurekekin · 2026-10-10