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

ElevenLabs alternatives

Updated · refreshed weekly · Methodology

Best model from each provider, by score. 121 models scored in all.
ModelScoreof 100Price$ / 1M charsCovalp50 msSpekop50 msOpen weightsCompared with Eleven v4
Eleven v4ElevenLabs · the reference86$80.00—844 msNoRanked #2 of 121. Latency on Speko only.
Simba 3.2SpeechifyAI87$10.00101 ms85 msNoAbout 8× cheaper. Scores 1 point above Eleven v4.
Sonic 3.6Cartesia83$49.00288 ms120 msNoWithin 3 points of Eleven v4. 7× faster on Speko (120 vs 844 ms).
Gemini 3.8 Flash TTSGoogle82$16.49—1031 msNo79% cheaper. Only on 2 of 3 boards, so the score is less certain.
Lightning V3.1 Pro (Oct 2026)Smallest.ai78$19.50——No76% cheaper. Only on 2 of 3 boards, so the score is less certain.
Fish Audio Drama 3Fish Audio74$15.00——No81% cheaper. Only on 2 of 3 boards, so the score is less certain.
Realtime TTS-2Inworld70$20.83161 ms116 msNo74% cheaper. 7.3× faster on Speko (116 vs 844 ms).
StepAudio 3 TTSStepFun69$14.00——No83% cheaper. A big drop in quality: 17 points behind Eleven v4.
Qwen-Audio-3.1-TTS-PlusAlibaba68$19.31——No76% cheaper. A big drop in quality: 18 points behind Eleven v4.
Voxtral 3 TTSMistral67$13.00——No84% cheaper. A big drop in quality: 19 points behind Eleven v4.
Speech 2.8 HDMiniMax63$100.00—294 msNo2.9× faster on Speko (294 vs 844 ms). Costs 25% more.
Breeze TTS 2BreezeBlue61$34.00——No57% cheaper. A big drop in quality: 25 points behind Eleven v4.
AssemblyAI-TTS-1AssemblyAI58$15.00204 ms—No81% cheaper. A big drop in quality: 28 points behind Eleven v4.
Falcon 2Murf AI57$10.00141 ms—NoAbout 8× cheaper. A big drop in quality: 29 points behind Eleven v4.
Luna TTSVUI Labs56$80.00——NoA big drop in quality: 30 points behind Eleven v4. Only on 1 of 3 boards, so the score is less certain.

One row per provider: its highest-scoring model. Latency is time to first audio, each source in its own column; "—" means that source has not published a figure. See every model.

Who beats Eleven v4, and who costs less

1 provider has a higher-scoring model than Eleven v4 (86): Simba 3.2 (87).

3 more land within 10 points: Sonic 3.6 (83), Gemini 3.8 Flash TTS (82), Lightning V3.1 Pro (Oct 2026) (78). All list below Eleven v4's $80.00 per 1M characters.

Of the models that score 60 or more, the cheapest outside ElevenLabs is Simba 3.2 by SpeechifyAI at $10.00 per 1M characters (score 87). That is 8× cheaper than Eleven v4.

Below $10 per 1M characters, the best score on the board is 33 (Gemini 2.5 Flash Lite TTS, $9.20). Under that line you pay with quality.

Open weights

The best open-weights model on the board is Fish Audio S2 Pro by Fish Audio: score 41, ranked #45, 45 points behind Eleven v4. You can host it yourself, so there is no per-character fee, but you pay for the GPU.

What this table does not show

7 of the 14 alternatives have no public latency figure at all (Smallest.ai, Fish Audio, StepFun, Alibaba, Mistral and others), so you cannot compare their speed here. The score comes from public listening arenas, and one of the three is US English only. It says little about other languages, voice cloning or SSML control. Those need your own test.

Questions people ask

What is the best ElevenLabs alternative?
Simba 3.2 by SpeechifyAI has the highest score of any non-ElevenLabs provider on this board: 87 against 86 for Eleven v4, at $10.00 per 1M characters. Next is Sonic 3.6 (Cartesia) at 83.
Is there a cheaper alternative to ElevenLabs with similar quality?
Sonic 3.6 (83), Gemini 3.8 Flash TTS (82), Lightning V3.1 Pro (Oct 2026) (78) score within 10 points of Eleven v4 and list lower. Of the models that score 60 or more, the cheapest outside ElevenLabs is Simba 3.2 by SpeechifyAI at $10.00 per 1M characters (score 87). That is 8× cheaper than Eleven v4. One of the three boards behind the score is US English only, so check your own language before you switch.
Is there an open-source ElevenLabs alternative?
The best open-weights model on the board is Fish Audio S2 Pro by Fish Audio: score 41, ranked #45, 45 points behind Eleven v4. You can host it yourself, so there is no per-character fee, but you pay for the GPU. The full list is on the open-source page.

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. Latency is time to first audio (the wait before sound starts), copied from Coval and Speko and never averaged across them. We do not measure anything ourselves. How latency is sourced.