# texttospeech.com — Full Reference

> The independent meta-leaderboard for text-to-speech APIs. We aggregate every public TTS benchmark (artificialanalysis.com, voicearena.com and humannessindex.vapi.ai) into one 0–100 quality score using Bayesian pairwise rank aggregation, list each model's price per 1M characters, and show time-to-first-audio latency from public latency benchmarks. This document contains the complete standings, a price-ranked table, per-source latency tables, all providers, the full methodology, the blog archive, and data provenance — everything needed to answer questions about the best, cheapest, fastest, or best-value text-to-speech API or model.

Last updated: Oct 11, 2026 · Snapshot: 2026-10-11 · 110 models · 44 providers · 3 boards

---

## Summary

Simba 3.2 by SpeechifyAI is currently the best text-to-speech API, scoring 89/100 across all 3 public TTS benchmarks at $10.00 per 1M characters. We rank 110 models from 44 providers by aggregating artificialanalysis.com, voicearena.com and humannessindex.vapi.ai into one score — a fixed formula, no editorial adjustments, no vendor weighting. The cheapest scored model is Kokoro 82M v1.0 at $0.70 per 1M characters.

---

## Direct answers

### What is the best text-to-speech API right now?

Simba 3.2 by SpeechifyAI is the top-ranked text-to-speech model on the texttospeech.com meta-leaderboard, scoring 89/100 with med confidence as of Oct 11, 2026. The score aggregates 3 independent public benchmarks — artificialanalysis.com, voicearena.com and humannessindex.vapi.ai — so it reflects agreement between boards rather than one board's view. It is priced at $10.00 per 1M characters. The ranking is recomputed weekly and the full standings are on the leaderboard.

### What is the cheapest text-to-speech API?

Kokoro 82M v1.0 by Kokoro is the cheapest scored text-to-speech model at $0.70 per 1M characters, ranked #55 of 110 with a score of 31/100. The cheapest model inside the top ten is Simba 3.2 by SpeechifyAI at $10.00 per 1M characters, so that is the cheapest option we would call competitive on quality. Prices are $ per 1M characters; per-minute pricing is converted to the same unit so every row compares like with like.

### What is the best open-weights text-to-speech model?

Fish Audio S2 Pro by Fish Audio is the highest-ranked open-weights model, #33 overall with 48/100 as of Oct 11, 2026. Open-weights models can be self-hosted instead of called through a vendor API, so their running cost depends on your own hardware rather than a per-character rate. The leaderboard can be filtered to open weights only.

### What is the fastest text-to-speech API?

On Coval's time to first audio (p50, us-east-1, 7-day window to Oct 11, 2026), the fastest model on this board is Realtime TTS-2 Flash by Inworld at 61 ms, followed by Simba 3.0 by SpeechifyAI (99 ms) and Simba 3.2 by SpeechifyAI (108 ms). Among the top ten by quality score, the fastest is Simba 3.2 by SpeechifyAI at 108 ms (#1 overall). Other public benchmarks measure with different harnesses, regions and sample sizes, so their numbers are shown beside Coval's, never averaged into it: on Speko's time to first audio (p50, US East) it is Realtime TTS-2 Flash by Inworld at 82 ms; on Humanness Index's time to first audio (median, measured June 2026) it is Sonic by Cartesia at 116 ms. Latency is shown separately and is not part of the 0–100 quality score. We do not measure latency ourselves; 39 of 110 models have a public figure, and the rest show a dash.

### How does texttospeech.com score text-to-speech models?

Every score is a Bayesian pairwise rank aggregation (BPRA) over 3 public boards. Each board contributes a model's rank as a fraction — #1 is 1.0, scaled by how many models the board ranks — a Bayesian coverage term pulls models that appear on few boards toward the field average, and an ecosystem-normalized Borda count turns the result into a 0–100 score: #1 on every board is exactly 100, and missing a board caps the ceiling. Ties break by Schulze path strength. We do not run our own listening tests; we aggregate boards that do, and publish the formula and every source snapshot.

### How often is the text-to-speech leaderboard updated?

The source boards are polled weekly and the ranking recomputes whenever any of them changes. The current snapshot was captured Oct 11, 2026. Every snapshot is archived and downloadable as JSON, so any past ranking can be reproduced from the same data and the same published formula.

### Is the ranking sponsored or influenced by vendors?

No. There is no vendor sponsorship, referral fee, or paid placement, and no code path that can elevate or demote a named vendor: every model goes through the identical formula, the formula is open source, and each published ranking links the exact source snapshots it was computed from. The source boards are also independent of the vendors they rank — that is one of the criteria a board must meet to be aggregated at all.

---

## Standings

110 models across 3 boards, ranked by Bayesian pairwise rank aggregation. Ranks #1–20 are the ranked standings; #21 onward are honorable mentions (BPRA-scored but outside the top 20).

| Rank | Model | Provider | Score | Confidence | Price | Coverage |
| --- | --- | --- | --- | --- | --- | --- |
| #1 | Simba 3.2 | SpeechifyAI | 89.0 | Med (86%) | $10.0 | 100% |
| #2 | Eleven v4 | ElevenLabs | 84.0 | High (77%) | $80.0 | 67% |
| #3 | Sonic 3.6 | Cartesia | 82.0 | High (76%) | $49.0 | 67% |
| #4 | Gemini 3.8 Flash-Lite TTS | Google | 81.0 | High (75%) | $11.0 | 67% |
| #5 | Gemini 3.8 Flash TTS | Google | 81.0 | High (74%) | $16.5 | 67% |
| #6 | Eleven v3 | ElevenLabs | 78.0 | Low (70%) | $100.0 | 100% |
| #7 | Sonic 3.5 | Cartesia | 76.0 | Med (81%) | $49.0 | 100% |
| #8 | Gemini 3.1 Flash TTS | Google | 74.0 | High (76%) | $18.3 | 67% |
| #9 | Eleven v4 Turbo | ElevenLabs | 71.0 | Low (62%) | $40.0 | 33% |
| #10 | Qwen-Audio-3.1-TTS-Plus | Alibaba | 69.0 | Low (62%) | $19.3 | 33% |
| #11 | Speech 2.8 HD | MiniMax | 69.0 | High (77%) | $100.0 | 67% |
| #12 | Fish Audio S2.1 Pro | Fish Audio | 68.0 | Med (84%) | $15.0 | 100% |
| #13 | Qwen-Audio-3.0-TTS-Plus | Alibaba | 67.0 | Low (62%) | $27.6 | 33% |
| #14 | Realtime TTS-2 | Inworld | 66.0 | Low (62%) | $20.8 | 33% |
| #15 | Breeze TTS 2 | BreezeBlue | 64.0 | Low (62%) | $34.0 | 33% |
| #16 | Luna TTS | VUI Labs | 63.0 | Low (62%) | $80.0 | 33% |
| #17 | Realtime TTS-2 Flash | Inworld | 63.0 | Low (62%) | $10.4 | 33% |
| #18 | StepAudio 2.5 TTS (Aug 2026) | StepFun | 61.0 | Low (62%) | $85.0 | 33% |
| #19 | Falcon 2 | Murf AI | 60.0 | Med (65%) | $10.0 | 67% |
| #20 | v3 Conversational | ElevenLabs | 60.0 | Low (62%) | $50.0 | 33% |
| #21 | Lightning V3.1 Pro (Jul 2026) | Smallest.ai | — | Low (54%) | $19.5 | 67% |
| #22 | Soniox TTS Real-Time v2 | Soniox | — | Low (62%) | $14.2 | 33% |
| #23 | Speech-02-HD | MiniMax | — | High (74%) | $100.0 | 67% |
| #24 | Gradium TTS (Sep 2026) | Gradium | — | Low (62%) | $47.8 | 33% |
| #25 | Speech 2.8 Turbo | MiniMax | — | Low (62%) | $60.0 | 33% |
| #26 | Gradium TTS (Aug 2026) | Gradium | — | Low (62%) | $47.2 | 33% |
| #27 | SpaceXAI TTS | xAI | — | Low (62%) | $15.0 | 33% |
| #28 | Step TTS 2 (Mar 2026) | StepFun | — | Low (62%) | $40.0 | 33% |
| #29 | Async Flash v1.5 | async | — | Low (62%) | $10.1 | 33% |
| #30 | Azure HD 2.5 | Microsoft | — | Low (62%) | $22.0 | 33% |
| #31 | Speech 2.6 HD | MiniMax | — | Low (62%) | $100.0 | 33% |
| #32 | Simba 3.0 | SpeechifyAI | — | Low (62%) | $12.5 | 33% |
| #33 | Fish Audio S2 Pro | Fish Audio | — | Low (58%) | $15.0 | 67% |
| #34 | Speech 2.6 Turbo | MiniMax | — | Low (62%) | $60.0 | 33% |
| #35 | Async Pro v1.0 | async | — | Low (62%) | $20.2 | 33% |
| #36 | Speech-02-Turbo | MiniMax | — | High (71%) | $60.0 | 67% |
| #37 | Speech 2.6 | Hithink | — | Med (63%) | — | 67% |
| #38 | TTS-1 HD | OpenAI | — | Low (62%) | $30.0 | 33% |
| #39 | Chatterbox HD | Resemble AI | — | Low (62%) | $40.0 | 33% |
| #40 | Turbo v2.5 | ElevenLabs | — | Low (62%) | $50.0 | 33% |
| #41 | Step Audio EditX (Mar 2026) | StepFun | — | Low (62%) | — | 33% |
| #42 | Multilingual v2 | ElevenLabs | — | Low (62%) | $100.0 | 33% |
| #43 | Flash v2.5 | ElevenLabs | — | High (72%) | $50.0 | 67% |
| #44 | TTS-1 | OpenAI | — | Low (62%) | $15.0 | 33% |
| #45 | Voxtral TTS | Mistral | — | Low (62%) | $16.0 | 33% |
| #46 | Gemini 2.5 Flash Lite TTS | Google | — | Low (62%) | $9.2 | 33% |
| #47 | OpenAudio S1 | Fish Audio | — | Low (62%) | $15.0 | 33% |
| #48 | Studio | Google | — | Low (62%) | $160.0 | 33% |
| #49 | GPT-Realtime-2 | OpenAI | — | Low (62%) | $191.6 | 33% |
| #50 | Journey | Google | — | Low (62%) | $160.0 | 33% |
| #51 | Maya 2 Flash | Maya Research | — | Low (62%) | — | 33% |
| #52 | Sonic 3 | Cartesia | — | Med (80%) | $49.0 | 100% |
| #53 | Polly Generative | Amazon | — | Low (62%) | $30.0 | 33% |
| #54 | T2A-01-HD | MiniMax | — | Low (62%) | $50.0 | 33% |
| #55 | Kokoro 82M v1.0 | Kokoro | — | Low (62%) | $0.7 | 33% |
| #56 | Magpie-Multilingual 357M (Feb 2026) | NVIDIA | — | Low (62%) | — | 33% |
| #57 | Chirp 3: HD | Google | — | Low (62%) | $30.0 | 33% |
| #58 | Gemini 2.5 Flash TTS (Dec 2025) | Google | — | Low (62%) | $9.2 | 33% |
| #59 | Coda | Rime | — | Low (62%) | $50.0 | 33% |
| #60 | Maya 2 Global | Maya Research | — | High (78%) | — | 67% |
| #61 | MiMo-V2.5-TTS | Xiaomi | — | Low (62%) | — | 33% |
| #62 | Flux TTS | Deepgram | — | Low (62%) | $45.0 | 33% |
| #63 | Async Flash v1.0 | async | — | Low (62%) | $10.1 | 33% |
| #64 | Octave 2 | Hume AI | — | Low (62%) | $87.5 | 33% |
| #65 | Maya1 | Maya Research | — | Low (62%) | — | 33% |
| #66 | OpenAudio S1 Mini | Fish Audio | — | Low (62%) | $15.0 | 33% |
| #67 | Octave TTS | Hume AI | — | Low (62%) | $87.5 | 33% |
| #68 | Higgs Audio V3 TTS | Boson AI | — | Low (62%) | $15.0 | 33% |
| #69 | Grok TTS | xAI | — | Med (63%) | $15.0 | 67% |
| #70 | Azure Neural | Microsoft | — | Low (62%) | $15.0 | 33% |
| #71 | Bland Speech v3 | Bland AI | — | Low (62%) | $40.0 | 33% |
| #72 | MAI-Voice-1 | Microsoft | — | Low (62%) | — | 33% |
| #73 | Gemini 2.5 Pro (Dec 2025) | Google | — | Low (62%) | $18.3 | 33% |
| #74 | Polly Long-Form | Amazon | — | Low (62%) | $100.0 | 33% |
| #75 | Lightning v3.1 | Smallest.ai | — | High (78%) | $25.0 | 67% |
| #76 | T2A-01-Turbo | MiniMax | — | Low (62%) | $30.0 | 33% |
| #77 | Chatterbox | Resemble AI | — | Low (62%) | $25.0 | 33% |
| #78 | MiMo-V2-TTS | Xiaomi | — | Low (62%) | — | 33% |
| #79 | Raon SpeechLM | Krafton | — | Low (62%) | — | 33% |
| #80 | Zonos-v0.1 | Zyphra | — | Low (62%) | $20.0 | 33% |
| #81 | Murf Speech Gen 2 | Murf AI | — | Low (62%) | $100.0 | 33% |
| #82 | Orpheus | Canopy Labs | — | Low (62%) | — | 33% |
| #83 | Realtime TTS-2 (Research Preview) | Inworld | — | Low (62%) | $20.8 | 33% |
| #84 | VibeVoice 1.5B | Microsoft | — | Low (62%) | — | 33% |
| #85 | Grok TTS (Streaming) | xAI | — | Low (62%) | $15.0 | 33% |
| #86 | OpenVoice v2 | OpenVoice | — | Low (62%) | $8.3 | 33% |
| #87 | Qwen3 TTS Flash | Alibaba | — | Low (62%) | $10.0 | 33% |
| #88 | Neuphonic TTS | Neuphonic | — | High (78%) | $20.8 | 67% |
| #89 | Eleven v3 Conversational | ElevenLabs | — | Low (62%) | — | 33% |
| #90 | Realtime TTS 1.5 Max | Inworld | — | Low (62%) | $26.0 | 33% |
| #91 | Gradium TTS | Gradium | — | Low (62%) | $47.2 | 33% |
| #92 | Kugel 3 | KugelAudio | — | Low (62%) | — | 33% |
| #93 | Turbo v2 | ElevenLabs | — | Low (62%) | — | 33% |
| #94 | Qwen3 TTS | Alibaba | — | Low (62%) | $13.8 | 33% |
| #95 | WaveNet | Google | — | Low (62%) | $16.0 | 33% |
| #96 | TTS-2 | Inworld | — | Low (62%) | — | 33% |
| #97 | XTTS v2 | Coqui | — | Low (62%) | $40.4 | 33% |
| #98 | Mist V2 | Rime | — | Low (62%) | $31.1 | 33% |
| #99 | Sonic 2 | Cartesia | — | Low (62%) | — | 33% |
| #100 | Dragon HD Omni | Microsoft | — | Low (62%) | — | 33% |
| #101 | Polly Neural | Amazon | — | Low (62%) | $16.0 | 33% |
| #102 | Neural2 | Google | — | Low (62%) | $16.0 | 33% |
| #103 | StyleTTS 2 | StyleTTS | — | Low (62%) | $2.8 | 33% |
| #104 | Standard | Google | — | Low (62%) | $4.0 | 33% |
| #105 | gpt-4o-mini-tts | OpenAI | — | Low (62%) | — | 33% |
| #106 | Noiz TTS | Noiz | — | Low (62%) | — | 33% |
| #107 | Gradium TTS | Gradium | — | Low (62%) | $47.2 | 33% |
| #108 | MetaVoice v1 | MetaVoice | — | Low (62%) | — | 33% |
| #109 | Sonic | Cartesia | — | Low (62%) | — | 33% |
| #110 | Polly Standard | Amazon | — | Low (62%) | $4.0 | 33% |

---

## Cheapest text-to-speech APIs (price-ranked)

All 90 scored models that publish a price, cheapest first. Price is $ per 1M characters; per-minute pricing is converted to the same unit (see price transparency below). The score sits beside each price deliberately: cheapest overall and cheapest worth using are rarely the same row.

| $/1M chars | Model | Provider | Score | Rank |
| --- | --- | --- | --- | --- |
| $0.70 | Kokoro 82M v1.0 | Kokoro | 31/100 | #55 |
| $2.80 | StyleTTS 2 | StyleTTS | 3/100 | #103 |
| $4.00 | Standard | Google | 2/100 | #104 |
| $4.00 | Polly Standard | Amazon | 0/100 | #110 |
| $8.30 | OpenVoice v2 | OpenVoice | 10/100 | #86 |
| $9.20 | Gemini 2.5 Flash Lite TTS | Google | 38/100 | #46 |
| $9.20 | Gemini 2.5 Flash TTS (Dec 2025) | Google | 28/100 | #58 |
| $10.00 | Simba 3.2 | SpeechifyAI | 89/100 | #1 |
| $10.00 | Falcon 2 | Murf AI | 60/100 | #19 |
| $10.00 | Qwen3 TTS Flash | Alibaba | 10/100 | #87 |
| $10.10 | Async Flash v1.5 | async | 51/100 | #29 |
| $10.10 | Async Flash v1.0 | async | 25/100 | #63 |
| $10.40 | Realtime TTS-2 Flash | Inworld | 63/100 | #17 |
| $11.00 | Gemini 3.8 Flash-Lite TTS | Google | 81/100 | #4 |
| $12.50 | Simba 3.0 | SpeechifyAI | 49/100 | #32 |
| $13.80 | Qwen3 TTS | Alibaba | 7/100 | #94 |
| $14.20 | Soniox TTS Real-Time v2 | Soniox | 59/100 | #22 |
| $15.00 | Fish Audio S2.1 Pro | Fish Audio | 68/100 | #12 |
| $15.00 | SpaceXAI TTS | xAI | 53/100 | #27 |
| $15.00 | Fish Audio S2 Pro | Fish Audio | 48/100 | #33 |
| $15.00 | TTS-1 | OpenAI | 40/100 | #44 |
| $15.00 | OpenAudio S1 | Fish Audio | 37/100 | #47 |
| $15.00 | OpenAudio S1 Mini | Fish Audio | 23/100 | #66 |
| $15.00 | Higgs Audio V3 TTS | Boson AI | 21/100 | #68 |
| $15.00 | Grok TTS | xAI | 20/100 | #69 |
| $15.00 | Azure Neural | Microsoft | 20/100 | #70 |
| $15.00 | Grok TTS (Streaming) | xAI | 10/100 | #85 |
| $16.00 | Voxtral TTS | Mistral | 39/100 | #45 |
| $16.00 | MAI-Voice-1 | Microsoft | 19/100 | #72 |
| $16.00 | WaveNet | Google | 7/100 | #95 |
| $16.00 | Polly Neural | Amazon | 4/100 | #101 |
| $16.00 | Neural2 | Google | 4/100 | #102 |
| $16.50 | Gemini 3.8 Flash TTS | Google | 81/100 | #5 |
| $18.30 | Gemini 3.1 Flash TTS | Google | 74/100 | #8 |
| $18.30 | Gemini 2.5 Pro (Dec 2025) | Google | 18/100 | #73 |
| $19.30 | Qwen-Audio-3.1-TTS-Plus | Alibaba | 69/100 | #10 |
| $19.50 | Lightning V3.1 Pro (Jul 2026) | Smallest.ai | 59/100 | #21 |
| $20.00 | Zonos-v0.1 | Zyphra | 13/100 | #80 |
| $20.20 | Async Pro v1.0 | async | 47/100 | #35 |
| $20.80 | Realtime TTS-2 | Inworld | 66/100 | #14 |
| $20.80 | Realtime TTS-2 (Research Preview) | Inworld | 11/100 | #83 |
| $20.80 | Neuphonic TTS | Neuphonic | 10/100 | #88 |
| $22.00 | Azure HD 2.5 | Microsoft | 50/100 | #30 |
| $22.00 | Dragon HD Omni | Microsoft | 4/100 | #100 |
| $25.00 | Lightning v3.1 | Smallest.ai | 17/100 | #75 |
| $25.00 | Chatterbox | Resemble AI | 16/100 | #77 |
| $26.00 | Realtime TTS 1.5 Max | Inworld | 8/100 | #90 |
| $27.60 | Qwen-Audio-3.0-TTS-Plus | Alibaba | 67/100 | #13 |
| $30.00 | TTS-1 HD | OpenAI | 45/100 | #38 |
| $30.00 | Polly Generative | Amazon | 33/100 | #53 |
| $30.00 | Chirp 3: HD | Google | 29/100 | #57 |
| $30.00 | T2A-01-Turbo | MiniMax | 16/100 | #76 |
| $31.10 | Mist V2 | Rime | 5/100 | #98 |
| $34.00 | Breeze TTS 2 | BreezeBlue | 64/100 | #15 |
| $40.00 | Eleven v4 Turbo | ElevenLabs | 71/100 | #9 |
| $40.00 | Step TTS 2 (Mar 2026) | StepFun | 51/100 | #28 |
| $40.00 | Chatterbox HD | Resemble AI | 44/100 | #39 |
| $40.00 | Bland Speech v3 | Bland AI | 19/100 | #71 |
| $40.40 | XTTS v2 | Coqui | 6/100 | #97 |
| $45.00 | Flux TTS | Deepgram | 26/100 | #62 |
| $47.20 | Gradium TTS (Aug 2026) | Gradium | 54/100 | #26 |
| $47.20 | Gradium TTS | Gradium | 8/100 | #91 |
| $47.20 | Gradium TTS | Gradium | 1/100 | #107 |
| $47.80 | Gradium TTS (Sep 2026) | Gradium | 56/100 | #24 |
| $49.00 | Sonic 3.6 | Cartesia | 82/100 | #3 |
| $49.00 | Sonic 3.5 | Cartesia | 76/100 | #7 |
| $49.00 | Sonic 3 | Cartesia | 33/100 | #52 |
| $50.00 | v3 Conversational | ElevenLabs | 60/100 | #20 |
| $50.00 | Turbo v2.5 | ElevenLabs | 43/100 | #40 |
| $50.00 | Flash v2.5 | ElevenLabs | 40/100 | #43 |
| $50.00 | T2A-01-HD | MiniMax | 32/100 | #54 |
| $50.00 | Coda | Rime | 28/100 | #59 |
| $60.00 | Speech 2.8 Turbo | MiniMax | 55/100 | #25 |
| $60.00 | Speech 2.6 Turbo | MiniMax | 48/100 | #34 |
| $60.00 | Speech-02-Turbo | MiniMax | 46/100 | #36 |
| $80.00 | Eleven v4 | ElevenLabs | 84/100 | #2 |
| $80.00 | Luna TTS | VUI Labs | 63/100 | #16 |
| $85.00 | StepAudio 2.5 TTS (Aug 2026) | StepFun | 61/100 | #18 |
| $87.50 | Octave 2 | Hume AI | 25/100 | #64 |
| $87.50 | Octave TTS | Hume AI | 22/100 | #67 |
| $100.00 | Eleven v3 | ElevenLabs | 78/100 | #6 |
| $100.00 | Speech 2.8 HD | MiniMax | 69/100 | #11 |
| $100.00 | Speech-02-HD | MiniMax | 56/100 | #23 |
| $100.00 | Speech 2.6 HD | MiniMax | 49/100 | #31 |
| $100.00 | Multilingual v2 | ElevenLabs | 41/100 | #42 |
| $100.00 | Polly Long-Form | Amazon | 17/100 | #74 |
| $100.00 | Murf Speech Gen 2 | Murf AI | 12/100 | #81 |
| $160.00 | Studio | Google | 37/100 | #48 |
| $160.00 | Journey | Google | 34/100 | #50 |
| $191.60 | GPT-Realtime-2 | OpenAI | 36/100 | #49 |

---

## Fastest text-to-speech APIs (latency)

Time to first audio in milliseconds, copied from public latency benchmarks — we never measure latency ourselves. The sources use different harnesses, regions and sample sizes, so each list below is one source's own numbers; they are never averaged. Coval is the primary source (continuously re-measured, largest sample). Latency is not part of the 0–100 score. 39 of 110 models have a public figure.

### Coval's time to first audio (p50, us-east-1, 7-day window to Oct 11, 2026)

Source: https://benchmarks.coval.ai/tts · Open-source harness run every 30 minutes from us-east-1; median over a rolling 7-day window. Streaming; connection handshakes excluded; counts leading silence until the first audible sample. · read Oct 11, 2026

- 61 ms — **Realtime TTS-2 Flash** (Inworld) — Score 63/100, ranked #17
- 99 ms — **Simba 3.0** (SpeechifyAI) — Score 49/100, ranked #32
- 108 ms — **Simba 3.2** (SpeechifyAI) — Score 89/100, ranked #1
- 149 ms — **Realtime TTS-2** (Inworld) — Score 66/100, ranked #14
- 149 ms — **TTS-2** (Inworld) — Score 6/100, ranked #96
- 187 ms — **Flash v2.5** (ElevenLabs) — Score 40/100, ranked #43
- 192 ms — **Eleven v4 Turbo** (ElevenLabs) — Score 71/100, ranked #9
- 259 ms — **Soniox TTS Real-Time v2** (Soniox) — Score 59/100, ranked #22
- 281 ms — **Sonic 3.5** (Cartesia) — Score 76/100, ranked #7
- 281 ms — **Falcon 2** (Murf AI) — Score 60/100, ranked #19
- 285 ms — **Sonic 3.6** (Cartesia) — Score 82/100, ranked #3
- 308 ms — **Fish Audio S2.1 Pro** (Fish Audio) — Score 68/100, ranked #12
- 320 ms — **v3 Conversational** (ElevenLabs) — Score 60/100, ranked #20
- 320 ms — **Eleven v3 Conversational** (ElevenLabs) — Score 9/100, ranked #89
- 335 ms — **Grok TTS** (xAI) — Score 20/100, ranked #69

### Speko's time to first audio (p50, US East)

Source: https://benchmarks.speko.ai/tts/ · Sequential requests from US East, about 30 per run, re-run per model on the date shown. · read Oct 11, 2026

- 82 ms (measured Sep 3, 2026) — **Realtime TTS-2 Flash** (Inworld) — Score 63/100, ranked #17
- 85 ms (measured Jul 3, 2026) — **Simba 3.2** (SpeechifyAI) — Score 89/100, ranked #1
- 105 ms (measured Oct 1, 2026) — **Gradium TTS** (Gradium) — Score 8/100, ranked #91
- 116 ms (measured Jul 3, 2026) — **Realtime TTS-2** (Inworld) — Score 66/100, ranked #14
- 116 ms (measured Jul 3, 2026) — **TTS-2** (Inworld) — Score 6/100, ranked #96
- 120 ms (measured Aug 28, 2026) — **Sonic 3.6** (Cartesia) — Score 82/100, ranked #3
- 121 ms (measured Jul 3, 2026) — **Sonic 3.5** (Cartesia) — Score 76/100, ranked #7
- 173 ms (measured Jul 3, 2026) — **Lightning v3.1** (Smallest.ai) — Score 17/100, ranked #75
- 185 ms (measured Jul 3, 2026) — **Fish Audio S2.1 Pro** (Fish Audio) — Score 68/100, ranked #12
- 205 ms (measured Sep 28, 2026) — **Eleven v4 Turbo** (ElevenLabs) — Score 71/100, ranked #9
- 266 ms (measured Jul 3, 2026) — **v3 Conversational** (ElevenLabs) — Score 60/100, ranked #20
- 266 ms (measured Jul 3, 2026) — **Eleven v3 Conversational** (ElevenLabs) — Score 9/100, ranked #89
- 272 ms (measured Jul 3, 2026) — **Grok TTS** (xAI) — Score 20/100, ranked #69
- 294 ms (measured Jul 3, 2026) — **Speech 2.8 HD** (MiniMax) — Score 69/100, ranked #11
- 381 ms (measured Aug 13, 2026) — **Soniox TTS Real-Time v2** (Soniox) — Score 59/100, ranked #22

### Humanness Index's time to first audio (median, measured June 2026)

Source: https://humannessindex.vapi.ai · Region not stated. 50 sequential streaming API requests per model, measured once (the date shown). · read Oct 11, 2026

- 116 ms — **Sonic** (Cartesia) — Score 0/100, ranked #109
- 128 ms — **Sonic 3.5** (Cartesia) — Score 76/100, ranked #7
- 141 ms — **Fish Audio S2.1 Pro** (Fish Audio) — Score 68/100, ranked #12
- 159 ms — **Sonic 2** (Cartesia) — Score 4/100, ranked #99
- 166 ms — **Sonic 3** (Cartesia) — Score 33/100, ranked #52
- 197 ms — **Flash v2.5** (ElevenLabs) — Score 40/100, ranked #43
- 276 ms — **Neuphonic TTS** (Neuphonic) — Score 10/100, ranked #88
- 285 ms — **Grok TTS (Streaming)** (xAI) — Score 10/100, ranked #85
- 288 ms — **TTS-2** (Inworld) — Score 6/100, ranked #96
- 302 ms — **Turbo v2** (ElevenLabs) — Score 7/100, ranked #93
- 315 ms — **Speech-02-Turbo** (MiniMax) — Score 46/100, ranked #36
- 325 ms — **Speech 2.8 HD** (MiniMax) — Score 69/100, ranked #11
- 332 ms — **Gradium TTS** (Gradium) — Score 1/100, ranked #107
- 337 ms — **Realtime TTS 1.5 Max** (Inworld) — Score 8/100, ranked #90
- 357 ms — **Speech-02-HD** (MiniMax) — Score 56/100, ranked #23

### Every model with a public latency figure

One column per source, never averaged; "—" = that source does not list the model.

| Rank | Model | Provider | Coval (ms) | Speko (ms) | Humanness Index (ms) |
| --- | --- | --- | --- | --- | --- |
| #1 | Simba 3.2 | SpeechifyAI | 108 | 85 | 428 |
| #2 | Eleven v4 | ElevenLabs | — | 844 | — |
| #3 | Sonic 3.6 | Cartesia | 285 | 120 | — |
| #4 | Gemini 3.8 Flash-Lite TTS | Google | — | 497 | — |
| #5 | Gemini 3.8 Flash TTS | Google | — | 1031 | — |
| #6 | Eleven v3 | ElevenLabs | — | — | 758 |
| #7 | Sonic 3.5 | Cartesia | 281 | 121 | 128 |
| #8 | Gemini 3.1 Flash TTS | Google | — | 978 | — |
| #9 | Eleven v4 Turbo | ElevenLabs | 192 | 205 | — |
| #11 | Speech 2.8 HD | MiniMax | — | 294 | 325 |
| #12 | Fish Audio S2.1 Pro | Fish Audio | 308 | 185 | 141 |
| #14 | Realtime TTS-2 | Inworld | 149 | 116 | — |
| #17 | Realtime TTS-2 Flash | Inworld | 61 | 82 | — |
| #19 | Falcon 2 | Murf AI | 281 | — | — |
| #20 | v3 Conversational | ElevenLabs | 320 | 266 | — |
| #21 | Lightning V3.1 Pro (Jul 2026) | Smallest.ai | 369 | — | — |
| #22 | Soniox TTS Real-Time v2 | Soniox | 259 | 381 | — |
| #23 | Speech-02-HD | MiniMax | — | — | 357 |
| #32 | Simba 3.0 | SpeechifyAI | 99 | — | — |
| #36 | Speech-02-Turbo | MiniMax | — | — | 315 |
| #43 | Flash v2.5 | ElevenLabs | 187 | — | 197 |
| #47 | OpenAudio S1 | Fish Audio | 382 | — | — |
| #52 | Sonic 3 | Cartesia | — | — | 166 |
| #57 | Chirp 3: HD | Google | 536 | — | — |
| #64 | Octave 2 | Hume AI | — | 448 | — |
| #69 | Grok TTS | xAI | 335 | 272 | 460 |
| #75 | Lightning v3.1 | Smallest.ai | — | 173 | 420 |
| #85 | Grok TTS (Streaming) | xAI | — | — | 285 |
| #87 | Qwen3 TTS Flash | Alibaba | 727 | 472 | — |
| #88 | Neuphonic TTS | Neuphonic | — | — | 276 |
| #89 | Eleven v3 Conversational | ElevenLabs | 320 | 266 | — |
| #90 | Realtime TTS 1.5 Max | Inworld | — | — | 337 |
| #91 | Gradium TTS | Gradium | — | 105 | — |
| #93 | Turbo v2 | ElevenLabs | — | — | 302 |
| #96 | TTS-2 | Inworld | 149 | 116 | 288 |
| #99 | Sonic 2 | Cartesia | — | — | 159 |
| #105 | gpt-4o-mini-tts | OpenAI | 2686 | 691 | — |
| #107 | Gradium TTS | Gradium | — | — | 332 |
| #109 | Sonic | Cartesia | — | — | 116 |

---

## Boards

- **artificialanalysis.com** (Speech Arena Elo) — 96 models ranked, status: live
- **voicearena.com** (arena Elo, US English) — 21 models ranked, status: live
- **humannessindex.vapi.ai** (humanness score 0–100) — 20 models ranked, status: live

---

## Providers

All 44 providers — those with models on any source board, plus recognized providers listed manually (no BPRA score).

- **Alibaba** — 4 models, on 1 board
- **Amazon** — 4 models, on 1 board
- **async** — 3 models, on 1 board
- **Bland AI** — 2 models, on 1 board
- **Boson AI** — 1 model, on 1 board
- **BreezeBlue** — 1 model, on 1 board
- **Canopy Labs** — 1 model, on 1 board
- **Cartesia** — 5 models, on 3 boards
- **Coqui** — 1 model, on 1 board
- **Deepgram** — 4 models, on 1 board
- **ElevenLabs** — 9 models, on 3 boards
- **Fish Audio** — 4 models, on 3 boards
- **Google** — 12 models, on 2 boards
- **Gradium** — 4 models, on 3 boards
- **Hithink** — 1 model, on 2 boards
- **Hume AI** — 2 models, on 1 board
- **Inworld** — 5 models, on 3 boards
- **Kokoro** — 1 model, on 1 board
- **Krafton** — 1 model, on 1 board
- **KugelAudio** — 1 model, on 1 board
- **Maya Research** — 3 models, on 2 boards
- **MetaVoice** — 1 model, on 1 board
- **Microsoft** — 5 models, on 2 boards
- **MiniMax** — 8 models, on 2 boards
- **Mistral** — 1 model, on 1 board
- **Murf AI** — 2 models, on 2 boards
- **Neuphonic** — 1 model, on 2 boards
- **Noiz** — 1 model, on 1 board
- **NVIDIA** — 1 model, on 1 board
- **OpenAI** — 4 models, on 2 boards
- **OpenVoice** — 1 model, on 1 board
- **Play.ht** — 3 recognized models (not scored — not on any board)
- **Resemble AI** — 2 models, on 1 board
- **Rime** — 2 models, on 1 board
- **Smallest.ai** — 2 models, on 3 boards
- **Soniox** — 1 model, on 1 board
- **SpeechifyAI** — 2 models, on 3 boards
- **StepFun** — 3 models, on 1 board
- **StyleTTS** — 1 model, on 1 board
- **VUI Labs** — 1 model, on 1 board
- **WellSaid Labs** — 1 recognized model (not scored — not on any board)
- **xAI** — 3 models, on 3 boards
- **Xiaomi** — 2 models, on 1 board
- **Zyphra** — 1 model, on 1 board

---

### Methodology: Bayesian Pairwise Rank Aggregation (BPRA)

Our meta-leaderboard runs a four-phase pipeline on the output of each source board, producing a single 0–100 score for every model that appears on at least one board.

**1. Fractional scaling** — Every board contributes a model's rank. Rank #1 is always 1.0; the last row is 1/N. Third place on a 1000-model board (>0.997) is worth more than third on a 10-model board (0.800). Formula: (N − R + 1) / N.

**2. Bayesian coverage smoothing** — A model on few boards is pulled toward the field average via a Bayesian prior, so a single strong placement cannot masquerade as a consensus. The pull shrinks as more boards cover the model.

**3. Ecosystem-normalized Borda count** — For each pair of models, we count how many competitors each defeats across every board. The total is divided by competitors across ALL 3 boards so that each board contributes equally to the available points. A model ranked #1 on every board scores exactly 100. Missing a board permanently caps the ceiling.

**4. Schulze tiebreak** — Ties in aggregate score are broken by the strongest chain of board-level wins between two models (the Schulze criterion). Handles circular preferences deterministically — no randomness.

**Confidence labels** — Every model carries a qualifier: High (≥2/3 boards with broad agreement), Med (≥2 boards), Low (1 board or disagreeing boards). Confidence is a recommendation of quality next to the score, never part of it.

**Latency** — Shown beside the score, never in it. Time to first audio is copied from public latency benchmarks (each figure linked, dated and kept under its own source, never averaged); we never measure it ourselves. Latency sources are best-effort: one being down never blocks the weekly refresh.

**Independence** — No vendor sponsorship, referral fees, or paid placement. Every engine goes through the identical formula. The formula and every source snapshot are public.

---

### Price transparency

Prices are listed as $X.XX per 1M characters. Some providers price per minute of audio; where a provider's listed price uses a per-minute format, we convert using a neutral benchmark — generate 1000 characters of audio through a reference model (Cartesia), measure the output duration in minutes, then derive the per-1M-characters equivalent. Converted prices are marked with an asterisk.
Prices are sourced from Artificial Analysis and providers' own pages.

---

## Blog

- [Sonic 3.6 (Cartesia) Review: Real-Time Conversational Leader, Long-Form Lagger](https://texttospeech.com/blog/sonic-3-6-review) (Sep 24, 2026, Reviews)
  Sonic 3.6 tops both Elo boards that measure it and sits #2 on the texttospeech.com aggregate, held off the top spot only by a missing Vapi Humanness Index entry. But version-specific evidence is thin, and the model's headline selling point, sub-90ms latency, does not survive independent measurement.

- [Simba 3.2 (Speechify) Review: Aggregate #1, Not a Unanimous Verdict](https://texttospeech.com/blog/simba-3-2-review) (Sep 17, 2026, Reviews)
  Simba 3.2 is the top-ranked model on the texttospeech.com aggregate, and the cheapest model at that quality tier. The #1 holds up as a value claim — but it is the output of our ranking methodology, not a unanimous verdict from the three boards that feed it.

- [Gemini 3.1 Flash TTS review: cheap and expressive, but broken past a minute](https://texttospeech.com/blog/gemini-3-1-flash-tts-review) (Sep 10, 2026, Reviews)
  Gemini 3.1 Flash TTS is one of the cheapest, most prompt-steerable text-to-speech models available, and for clips under a minute it is genuinely good. Past that, it has a confirmed, still-unfixed defect that makes it unsuitable for audiobooks or long narration.

- [ElevenLabs Eleven v3 review: the most expressive voice in TTS — and the slowest flagship](https://texttospeech.com/blog/eleven-v3-review) (Sep 3, 2026, Reviews)
  Eleven v3 is the most expressive text-to-speech model of its generation and, at 97/100, still the most human-sounding voice on Vapi's Humanness Index. That quality costs 758ms of latency, and through 2026 it slid from roughly #2 to #15 on Artificial Analysis as faster competitors shipped.

- [First look: the new TTS models that landed in August 2026](https://texttospeech.com/blog/new-tts-models-august-2026) (Aug 31, 2026, First look)
  Six new TTS models landed on the boards in August 2026. Sonic 3.6 topped both boards that rate it. The other five are single-board debuts our aggregate is still waiting to confirm.

- [Best TTS providers and models (August 2026)](https://texttospeech.com/blog/2026-08-13) (Aug 13, 2026, Rankings)
  The inaugural TTS meta-leaderboard snapshot. Here's where the board stands as of August 2026.

- [How the meta-leaderboard works: our methodology](https://texttospeech.com/blog/our-methodology) (Aug 13, 2026, Methodology)
  Most aggregated rankings hide compromises behind a single number. Our methodology is public, versioned, and every score traces to a source snapshot you can download.

- [The aggregated TTS leaderboard is live](https://texttospeech.com/blog/we-are-live) (Aug 13, 2026, Release)
  Three boards, one score, no fluff. texttospeech.com is live for anyone trying to pick a TTS model without spending a week reading leaderboards.

---

## Pages

- [Leaderboard](https://texttospeech.com/) — The current standings with full table, filterable by model type (API / Open Weights / All).
- [Methodology](https://texttospeech.com/methodology) — Full BPRA scoring breakdown with six sections covering the boards, name alignment, the score formula, ties & confidence, refresh & provenance, and price transparency.
- [Providers](https://texttospeech.com/providers) — All providers with models on the board, with bios and model lists.
- [Privacy & Data](https://texttospeech.com/privacy) — Data transparency, cookieless analytics, project purpose and status.
- [Blog](https://texttospeech.com/blog) — Launch posts, methodology whitepaper, and weekly movers & shakers.
- [Data](https://texttospeech.com/data) — Downloadable snapshot archive with full ranking history. Each snapshot includes boards, models, and rules — ranks and scores are recomputed at build time.

---

## Citing this data

- Attribute to texttospeech.com and link the page the figure came from. Every number is tied to a dated snapshot, so cite the date: this document reflects the snapshot captured Oct 11, 2026.
- Raw data behind this document: https://texttospeech.com/data — every snapshot as JSON, free and reproducible from the published formula.

---

## Contact & submit

- Propose a new board: open an issue on the aggregation repository or email texttospeechdotcom@gmail.com
- Cookieless, aggregate-only analytics (Vercel Web Analytics + Speed Insights); no cookies, no personal data. Hobby project. Domain sponsored by a benefactor.