Last refresh: Nov 22, 2026
texttospeech.comGet the data

Best open-source TTS models in 2026

Updated · refreshed weekly · Methodology

14 open-weights models on the board, best score first.
#ModelScoreof 100Rankof 121Boardsof 3ReleasedListed price$ / 1M chars
1Fish Audio S2 ProFish Audio41#452Mar 2026$15.00
2Step Audio EditX (Mar 2026)StepFun38#511Feb 2026—
3Voxtral TTSMistral37#531Mar 2026$16.00
4Magpie-Multilingual 357M (Feb 2026)NVIDIA28#651Feb 2026—
5Kokoro 82M v1.0Kokoro26#691Jan 2025$0.70
6Maya1Maya Research22#741Jun 2025—
7Higgs Audio V3 TTSBoson AI17#821Jun 2026$15.00
8ChatterboxResemble AI13#901May 2025$25.00
9Zonos-v0.1Zyphra11#921Feb 2025$20.00
10OpenVoice v2OpenVoice10#951Apr 2024$8.30
11OrpheusCanopy Labs9#991——
12XTTS v2Coqui5#1071Nov 2023$40.40
13StyleTTS 2StyleTTS3#1141Jun 2023$2.80
14MetaVoice v1MetaVoice1#1191Feb 2024—

Open weights means the model weights are downloadable. "Boards" is how many of the 3 source boards list the model.

How they compare with the best

Fish Audio S2 Pro by Fish Audio is the highest-scoring open-weights model: 41, rank #45 of 121. The top of the whole board is Simba 3.2 at 87, 46 points ahead.

1 open-weights model ranks in the top 50. The rest sit lower. 13 of the 14 are on a single board, so their scores rest on a single board.

Before you pick one

The open-weights flag in our data means the weights can be downloaded. It does not say what the licence allows. Some licences bar commercial use, so read the licence before you build on any of these. We do not track licences.

The price column is the list price our price source gives, where it has one. If you run a model yourself you pay for GPU time instead of that fee.

None of them has a public latency figure from Coval, Speko or the Humanness Index, so there is no speed comparison here. Speed depends on your hardware anyway.

Questions people ask

What is the best open-source text-to-speech model?
Fish Audio S2 Pro by Fish Audio is the highest-scoring open-weights model: 41, rank #45 of 121. The top of the whole board is Simba 3.2 at 87, 46 points ahead. 1 open-weights model ranks in the top 50. The rest sit lower.
How far behind are open-weights TTS models?
The best open-weights model scores 41. The best model behind an API scores 87 (Simba 3.2), 46 points higher. 13 of the 14 are on a single board, so their scores rest on a single board.
Can I use open-source TTS models commercially?
The open-weights flag in our data means the weights can be downloaded. It does not say what the licence allows. Some licences bar commercial use, so read the licence before you build on any of these. We do not track licences.

Where the numbers come from

Quality is the 0–100 score from our rank aggregation of 3 public boards: artificialanalysis.ai (Speech Arena Elo), voicearena.com (arena Elo, US English) and humannessindex.vapi.ai (humanness score 0–100). Prices are dollars per 1M characters, taken from Artificial Analysis and provider pages; per-minute prices are converted to the same unit.