At about 05:23 Beijing time on October 23, 2026, Artificial Analysis posted on X that Ant Group’s Ming-3.0-tts, associated with AntLingAGI, ranks at the top tier of its open text-to-speech boards while using only 0.5B parameters and an MIT license. DigIntoAI has not tested the model or the leaderboards.
Per Artificial Analysis, Ming-3.0-tts is the only model under 1B parameters in the top three of any of its text-to-speech leaderboards. On the Controlled Voice Arena open-weights top ten, Artificial Analysis says every other model is 3.5B or larger, so Ming-3.0-tts reaches the same tier with about 7× to 70× fewer parameters. Weights labeled Ming-3.0-tts-0.5B and Ming-3.0-tts-tiny are listed on Hugging Face under the inclusionAI account with an MIT license on the model cards.
DigIntoAI previously covered another open TTS stack in Gradium Phonon offline TTS; a Phonon versus Ming comparison remains watch-only for now. For related voice and open-weights notes, see Grok Voice 3.0 and Muse Spark open weights.
We have not run Ming-3.0-tts, have not reproduced Artificial Analysis’s TTS rankings, and have not verified the Hugging Face checkpoints on a live workload.