Ant Group's Ming-3.0-tts takes #1 open-weights spots on TTS leaderboards at 93.2% pronunciation accuracy
ArtificialAnlys · x · 2026-10-20
Artificial Analysis's latest TTS evaluation crowns Ant Group's Ling AGI team model Ming-3.0-tts as the top open-weights text-to-speech model.
- Specs: A dense 0.5B-parameter model with preset voices and stateless voice cloning, plus a 7.9B MoE variant (Ming-3.0-tts-tiny), released under MIT on Hugging Face.
- Pronunciation Robustness #1: 93.2%, tied with Fish Audio's Drama 3 Preview and ahead of Eleven v4 (91.7%); ranks #1 in Expanding Shorthand (96.5%) and Preserving Exact Sequences (90.6%).
- Controlled Voice Arena: Elo 1,069, #1 among open weights, #13 overall.
- Provider Voice Arena: Elo 1,188, #3 open weights and #19 overall, strongest in Customer Service and Assistants categories. It tops open-weights rankings on two of three TTS leaderboards.
More from Models
- What Gen Models Run Well on a 3060/4060? Reddit Weighs Local Setups — Haghiri75 · 2026-10-20
- Sol 6.1 caught cheerfully talking itself through task assignment in agent logs — TheZachMueller · 2026-10-20
- Unreleased Moonshot model Kimi-K3.1 spotted in sglang PR, specs unconfirmed — TheZachMueller · 2026-10-20
- Claude Opus 5.5 tops TxBench antibody-discovery benchmark at 67.7%, GPT-6 Astra leads oligo tasks — kenbwork · 2026-10-20
- stealthbench adds 10 EDR-guarded tasks: models' opsec behavior is abysmal — dyn___ · 2026-10-20
- "Everyone has a little Terry Tao in their pocket now" — Reza_Zadeh · 2026-10-20