Hayz Labs · https://hayzlabs.com/reviews/top/voice-models/
reviews · top 20 list · checked 18 Oct
Top 20 AI voice models
Models that turn text into natural speech, for narration, voice assistants and dubbing.
Nothing on this list is scored yet: no outside source we can turn into a number covers them. The order below follows the public measure named on each row.
Ordered by score on the Artificial Analysis Speech Arena, where people pick the more natural of two voices without knowing which model made them. Open the Table to see the number that placed each one and how it was scored.
- 1Eleven v4 TurboElevenLabs
Top-rated natural voice in blind tests at half Eleven v4's price, fast enough for live agents. Not the cheapest.
Best for Voice agents, apps and narration that need the most natural voice fastYou give up About twice the price of the 3rd and 5th ranked voices; trains on your content by default
Not scored yetFree - 2Eleven v4ElevenLabs
Top-rated expressive voice for audiobooks and characters. Eleven v4 Turbo ties it in blind tests at half the price.
Best for Audiobooks and dubbing with expressive character voicesYou give up Twice v4 Turbo's price for the same rank; trains on your content by default
Not scored yetFree - 3Qwen-Audio-3.1-TTS-PlusAlibaba
Third in blind votes, a strong pick for apps already on Alibaba Cloud
Best for Near-top voices at under half ElevenLabs' price per characterYou give up Name and price don't match any Alibaba page yet
Not scored yetNo public price - 4Sonic 3.6Cartesia
Real-time voice agents that need near-top voice quality
Best for Phone and voice agents in 44 languagesYou give up Eleven v4 Turbo rates higher and costs less
Not scored yetFree - 5Gemini 3.8 Flash TTSGoogle
Near-top AI voices for less than half the price of ElevenLabs, with a free tier. Prices double on 1 Jan 2027.
Best for Narration or read-aloud in apps at volume, in over 130 languagesYou give up Ranks below both Eleven v4 models; free tier inputs are used to improve Google products
Not scored yetFree - 6Inworld Realtime TTS-2Inworld
Voice agents and games that want top-ten quality at a mid price
Best for Expressive voices you can steer with plain language, for agents and gamesYou give up Gemini 3.8 Flash TTS ranks higher and costs less per character
Not scored yetFree - 7Gemini 3.8 Flash-Lite TTSGoogle
Top-ten text to speech for a third less than Gemini 3.8 Flash TTS, with a free tier; pick Eleven v4 for the best voices.
Best for Lots of natural speech, such as read-aloud or app voicesYou give up Prices double on 1 Jan 2027; Simba 3.2 ranks level for about 40% less
Not scored yetFree - 8Simba 3.2Speechify
Cheapest of the top ten per character, good for large narration jobs
Best for Large narration jobs, with 500K free characters a monthYou give up Ranks below the Gemini and Inworld voices above it; no review yet
Not scored yetFree - 9Fish Audio Drama 3 (Preview)Fish Audio
Highest-ranked open-weights voice on the board; free to self-host for smaller firms
Best for Smaller firms that want to self-host a voice in 83 languagesYou give up Free weights only under $500K yearly revenue; API rises to $50 per 1M in November
Not scored yetPay as you go - 10Breeze TTS 2BreezeBlue
Open weights for research use; commercial use needs BreezeBlue's paid hosted service
Best for Research use with open voice weightsYou give up Commercial use needs BreezeBlue's paid hosted service; no price in our files
Not scored yetNo public price - 11Inworld Realtime TTS-2 FlashInworld
Cheaper, faster Inworld voice for agents where cost matters more than polish
Best for Voice agents that need very fast replies at a low priceYou give up Gemini 3.8 Flash-Lite TTS ranks higher at about the same price
Not scored yetFree - 12Luna TTSVUI Labs
Newer Chinese voice model ranking mid-teens; check access before you plan around it
Best for Teams able to confirm access who want a mid-table voiceYou give up Access is unclear and no price is listed; not reviewed yet
Not scored yetNo public price - 13StepAudio 2.5 TTSStepFun
Expressive voice from StepFun, but among the priciest on the board per character
Best for Expressive voices where cost is a lesser concernYou give up Among the priciest per character on this board; not reviewed yet
Not scored yetNo public price - 14Fish Audio S2.1 ProFish Audio
Low-cost voice cloning and narration from Fish Audio's main line
Best for Low-cost voice cloning and narration, with a free tierYou give up Ranks 18th of 96, below the open Fish Audio Drama 3
Not scored yetFree - 15Soniox TTS Real-Time v2Soniox
Low-cost real-time speech for voice agents at about $0.70 per hour of audio
Best for Low-cost live speech for voice agentsYou give up Ranks 20th of 96, well behind the top tier; not reviewed yet
Not scored yetPay as you go - 16Gradium TTSGradium
Solid voices with a free monthly allowance; behind the top tier
Best for Trying voices with 45K free characters a monthYou give up Behind the top tier; paid plans start at $13 a month
Not scored yetFree - 17MiniMax Speech 2.8 HDMiniMax
Solid voice, but at $100 per 1M chars it costs more than most models ranked above it
Best for Higher-quality MiniMax voice for narration and appsYou give up At $100 per 1M characters, costs more than most voices ranked above it
Not scored yetPay as you go - 18Lightning V3.1 ProSmallest.ai
Mid-table voice from Smallest.ai; check its price before choosing over higher-ranked picks
Best for A mid-table voice from Smallest.ai for apps and agentsYou give up No price in our files to compare with higher-ranked picks
Not scored yetNo public price - 19Falcon 2Murf AI
Murf's newest voice model; good but noticeably behind the leaders in blind votes
Best for Voiceovers from Murf's newest voice modelYou give up Noticeably behind the leaders in blind votes; not reviewed yet
Not scored yetNo public price - 20MiniMax Speech 2.8 TurboMiniMax
Faster, cheaper MiniMax tier; most will find better value higher up this list
Best for A faster, cheaper MiniMax voice for appsYou give up Lowest rank on this list, at $60 per 1M characters
Not scored yetPay as you go
| # | Name | Score | What placed it | Price | Best for | You give up |
|---|---|---|---|---|---|---|
| 1 | Eleven v4 TurboElevenLabs · Closed · our review | Not scored | Ranked #1 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1329. AA Speech Arena #1 of 96 (all rows distinct models down to #25), range 1-2; no independent TTS eval to adjust with; matches review Report a mistake | API $0.04 per 1K chars ($40 per 1M); Starter $6/mo | Voice agents, apps and narration that need the most natural voice fast | About twice the price of the 3rd and 5th ranked voices; trains on your content by default |
| 2 | Eleven v4ElevenLabs · Closed · our review | Not scored | Ranked #2 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1328. AA Speech Arena #2 of 96 (all rows distinct models down to #25), range 1-2; no independent TTS eval to adjust with; matches review Report a mistake | API $0.08 per 1K chars ($80 per 1M); Starter $6/mo | Audiobooks and dubbing with expressive character voices | Twice v4 Turbo's price for the same rank; trains on your content by default |
| 3 | Qwen-Audio-3.1-TTS-PlusAlibaba · Closed | Not scored | Ranked #3 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1295. AA Speech Arena #3 of 96 (all rows distinct models down to #25), range 3-4; no independent TTS eval to adjust with Report a mistake | Not confirmed | Near-top voices at under half ElevenLabs' price per character | Name and price don't match any Alibaba page yet |
| 4 | Sonic 3.6Cartesia · Closed | Not scored | Ranked #4 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1286. AA Speech Arena #4 of 96 (all rows distinct models down to #25), range 3-5; range does not include #1; no independent TTS eval to adjust with Report a mistake | Free 20K chars/mo; Pro $5/mo (1 credit per character) | Phone and voice agents in 44 languages | Eleven v4 Turbo rates higher and costs less |
| 5 | Gemini 3.8 Flash TTSGoogle · Closed · our review | Not scored | Ranked #5 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1278. AA Speech Arena #5 of 96 (all rows distinct models down to #25), range 4-6; range does not include #1; matches review; no independent TTS eval to adjust with Report a mistake | Free tier; API $0.50 in (text) / $9.00 out (audio) per 1M tokens to 2026-12-31, doubles 2027 | Narration or read-aloud in apps at volume, in over 130 languages | Ranks below both Eleven v4 models; free tier inputs are used to improve Google products |
| 6 | Inworld Realtime TTS-2Inworld · Closed | Not scored | Ranked #7 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1262. AA Speech Arena #7 of 96 (all rows distinct models down to #25), range 6-7; no independent TTS eval to adjust with Report a mistake | Free 70 min; API $25 per 1M chars | Expressive voices you can steer with plain language, for agents and games | Gemini 3.8 Flash TTS ranks higher and costs less per character |
| 7 | Gemini 3.8 Flash-Lite TTSGoogle · Closed · our review | Not scored | Ranked #8 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1247. AA Speech Arena #8 of 96 (all rows distinct models down to #25), range 8-10; no independent TTS eval to adjust with; matches review Report a mistake | Free tier; API $0.50 in (text) / $6.00 out (audio) per 1M tokens to 2026-12-31, doubles 2027 | Lots of natural speech, such as read-aloud or app voices | Prices double on 1 Jan 2027; Simba 3.2 ranks level for about 40% less |
| 8 | Simba 3.2Speechify · Closed | Not scored | Ranked #9 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1245. AA Speech Arena #9 of 96 (all rows distinct models down to #25), range 8-10; no independent TTS eval to adjust with Report a mistake | Free 500K chars/mo; API $6 to $10 per 1M chars beyond plan | Large narration jobs, with 500K free characters a month | Ranks below the Gemini and Inworld voices above it; no review yet |
| 9 | Fish Audio Drama 3 (Preview)Fish Audio · Open weights | Not scored | Ranked #10 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1243. AA Speech Arena #10 of 96 (all rows distinct models down to #25), range 8-10; 1,991 samples, not marked preliminary; no independent TTS eval to adjust with Report a mistake | API $50 per 1M chars ($10 launch price to 2026-11-12); weights free under $500K revenue | Smaller firms that want to self-host a voice in 83 languages | Free weights only under $500K yearly revenue; API rises to $50 per 1M in November |
| 10 | Breeze TTS 2BreezeBlue · Open weights | Not scored | Ranked #11 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1221. AA Speech Arena #11 of 96 (all rows distinct models down to #25), range 11-16; no independent TTS eval to adjust with Report a mistake | Not confirmed | Research use with open voice weights | Commercial use needs BreezeBlue's paid hosted service; no price in our files |
| 11 | Inworld Realtime TTS-2 FlashInworld · Closed | Not scored | Ranked #12 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1221. AA Speech Arena #12 of 96 (all rows distinct models down to #25), range 11-16; no independent TTS eval to adjust with Report a mistake | Free 70 min; API $15 per 1M chars | Voice agents that need very fast replies at a low price | Gemini 3.8 Flash-Lite TTS ranks higher at about the same price |
| 12 | Luna TTSVUI Labs · Closed | Not scored | Ranked #13 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1216. AA Speech Arena #13 of 96 (all rows distinct models down to #25), range 11-16; no independent TTS eval to adjust with Report a mistake | Not confirmed | Teams able to confirm access who want a mid-table voice | Access is unclear and no price is listed; not reviewed yet |
| 13 | StepAudio 2.5 TTSStepFun · Closed | Not scored | Ranked #14 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1215. AA Speech Arena #14 of 96 (all rows distinct models down to #25), range 11-16; no independent TTS eval to adjust with Report a mistake | Not confirmed | Expressive voices where cost is a lesser concern | Among the priciest per character on this board; not reviewed yet |
| 14 | Fish Audio S2.1 ProFish Audio · Closed | Not scored | Ranked #18 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1190. AA Speech Arena #18 of 96 (all rows distinct models down to #25), range 17-20; no independent TTS eval to adjust with Report a mistake | API $15 per 1M UTF-8 bytes; free s2.1-pro-free tier | Low-cost voice cloning and narration, with a free tier | Ranks 18th of 96, below the open Fish Audio Drama 3 |
| 15 | Soniox TTS Real-Time v2Soniox · Closed | Not scored | Ranked #20 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1179. AA Speech Arena #20 of 96 (all rows distinct models down to #25), range 18-23; no independent TTS eval to adjust with Report a mistake | API about $0.70 per hour of generated speech | Low-cost live speech for voice agents | Ranks 20th of 96, well behind the top tier; not reviewed yet |
| 16 | Gradium TTSGradium · Closed | Not scored | Ranked #21 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1173. AA Speech Arena #21 of 96 (all rows distinct models down to #25), range 19-23; Sep 2026 version; no independent TTS eval to adjust with Report a mistake | Free 45K chars/mo; XS $13/mo (1 credit per character) | Trying voices with 45K free characters a month | Behind the top tier; paid plans start at $13 a month |
| 17 | MiniMax Speech 2.8 HDMiniMax · Closed | Not scored | Ranked #22 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1172. AA Speech Arena #22 of 96 (all rows distinct models down to #25), range 19-23; no independent TTS eval to adjust with Report a mistake | API $100 per 1M chars | Higher-quality MiniMax voice for narration and apps | At $100 per 1M characters, costs more than most voices ranked above it |
| 18 | Lightning V3.1 ProSmallest.ai · Closed | Not scored | Ranked #23 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1171. AA Speech Arena #23 of 96 (all rows distinct models down to #25), range 19-23; no independent TTS eval to adjust with Report a mistake | Not confirmed | A mid-table voice from Smallest.ai for apps and agents | No price in our files to compare with higher-ranked picks |
| 19 | Falcon 2Murf AI · Closed | Not scored | Ranked #24 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1155. AA Speech Arena #24 of 96 (all rows distinct models down to #25), range 24-27; no independent TTS eval to adjust with Report a mistake | Not confirmed | Voiceovers from Murf's newest voice model | Noticeably behind the leaders in blind votes; not reviewed yet |
| 20 | MiniMax Speech 2.8 TurboMiniMax · Closed | Not scored | Ranked #25 of 96 in a public blind vote for voices (Artificial Analysis)NotesTheir figure: AA Speech Arena Elo 1153. AA Speech Arena #25 of 96 (all rows distinct models down to #25), range 24-27; no independent TTS eval to adjust with Report a mistake | API $60 per 1M chars | A faster, cheaper MiniMax voice for apps | Lowest rank on this list, at $60 per 1M characters |
Nothing matches all of that. , or try everything.
How we ranked this
Order. Ordered by score on the Artificial Analysis Speech Arena, where people pick the more natural of two voices without knowing which model made them.
How we score this list. No score yet. This leaderboard's terms do not let us publish numbers made from its data, so the order is the score and we add none of our own. The full rules are on How we score. Open "How it was scored" on any row to see the working.
What we did not do. We did not run any of these ourselves. Every number comes from the public source linked on its row, checked on 18 Oct 2026.
Spot a mistake? A wrong rank, price or link: tell us and we will check it against the source. Corrections are dated in News.
Reviews are drafted with help from AI models, including Anthropic's Claude. Model scores come only from outside leaderboards and independent evaluations, by a written rule.