Lumina Bench · Source-native media intelligence
Provider-voice, controlled-voice, and speech-to-speech evidence without mixing incompatible scales.
Compare provider-selected voices using Artificial Analysis Arena Elo.
Compare models using the controlled-voice Arena and a shared eight-voice evaluation set.
Compare the published Speech-to-Speech Index, latency, cost, and benchmark components.
Latest evidence snapshot
14 Aug 2026
Comparable records
159 across pathways
Ranking policy
Source metrics stay separate
Visual roundup
A concise view of each pathway. Open a full leaderboard for filters, complete tables and source detail.
Source metrics remain separate
Compare provider-selected voices using Artificial Analysis Arena Elo.
Top 12 · Current models
Current text-to-speech leaders using the source-native Arena Elo.
| Model | Creator | Arena Elo | 95% CI | Samples | Highlighted |
|---|---|---|---|---|---|
| Qwen-Audio-3.0-TTS-Plus | Alibaba | 1229.0 | ±15.0 | 1,579 | No |
| Simba 3.2 | SpeechifyAI | 1227.0 | ±15.0 | 1,589 | No |
| Gemini 3.1 Flash TTS | 1210.0 | ±13.0 | 2,937 | No | |
| Luna TTS | VUI Labs | 1209.0 | ±15.0 | 1,552 | No |
| Sonic 3.5 | Cartesia | 1203.0 | ±13.0 | 2,419 | No |
| StepAudio 2.5 TTS | StepFun | 1201.0 | ±15.0 | 1,306 | No |
| Realtime TTS 1.5 Max | Inworld | 1194.0 | ±13.0 | 2,126 | No |
| Lightning V3.1 Pro (Jul 2026) | Smallest.ai | 1192.0 | ±16.0 | 1,227 | No |
| Realtime TTS-2 - Research Preview | Inworld | 1191.0 | ±13.0 | 2,148 | No |
| Speech 2.8 HD | MiniMax | 1172.0 | ±12.0 | 4,017 | No |
| Eleven v3 | ElevenLabs | 1171.0 | ±12.0 | 3,890 | No |
| Async Flash v1.5 | async | 1151.0 | ±14.0 | 1,633 | No |
USD per 1M characters
Price is decision context and never affects the source ranking.
| Model · Provider | Creator | Cost | Arena Elo | Highlighted |
|---|---|---|---|---|
| Qwen-Audio-3.0-TTS-Plus · Alibaba | Alibaba | $27.6 /1M characters | 1229.0 | No |
| Simba 3.2 · SpeechifyAI | SpeechifyAI | $10.0 /1M characters | 1227.0 | No |
| Gemini 3.1 Flash TTS · Google | $18.3 /1M characters | 1210.0 | No | |
| Luna TTS · VUI Labs | VUI Labs | $80.0 /1M characters | 1209.0 | No |
| Sonic 3.5 · Cartesia | Cartesia | $49.0 /1M characters | 1203.0 | No |
| StepAudio 2.5 TTS · StepFun | StepFun | $85.0 /1M characters | 1201.0 | No |
| Realtime TTS 1.5 Max · Inworld | Inworld | $26.0 /1M characters | 1194.0 | No |
| Lightning V3.1 Pro (Jul 2026) · Smallest.ai | Smallest.ai | $19.5 /1M characters | 1192.0 | No |
| Realtime TTS-2 - Research Preview · Inworld | Inworld | $20.8 /1M characters | 1191.0 | No |
| Speech 2.8 HD · MiniMax | MiniMax | $100 /1M characters | 1172.0 | No |
| Eleven v3 · ElevenLabs | ElevenLabs | $100 /1M characters | 1171.0 | No |
| Async Flash v1.5 · async | async | $10.1 /1M characters | 1151.0 | No |
| Speech 2.8 Turbo · MiniMax | MiniMax | $60.0 /1M characters | 1146.0 | No |
| Step TTS 2 (Mar 2026) · StepFun | StepFun | $40.0 /1M characters | 1139.0 | No |
| Fish Audio S2.1 Pro · Fish Audio | Fish Audio | $15.0 /1M characters | 1138.0 | No |
| Lightning V3.1 Pro TTS (Jun 2026) · Smallest.ai | Smallest.ai | $19.5 /1M characters | 1138.0 | No |
| Speech 2.6 HD · MiniMax | MiniMax | $100 /1M characters | 1134.0 | No |
| Realtime TTS 1.5 Mini · Inworld | Inworld | $10.4 /1M characters | 1131.0 | No |
| Async Pro v1.0 · async | async | $20.2 /1M characters | 1124.0 | No |
| Speech 2.6 Turbo · MiniMax | MiniMax | $60.0 /1M characters | 1124.0 | No |
| Azure HD 2.5 · Microsoft | Microsoft | $22.0 /1M characters | 1122.0 | No |
| Simba 3.0 · SpeechifyAI | SpeechifyAI | $12.5 /1M characters | 1122.0 | No |
| Fish Audio S2 Pro · Fish Audio | Fish Audio | $15.0 /1M characters | 1121.0 | No |
| SpaceXAI TTS · SpaceXAI | SpaceXAI | $15.0 /1M characters | 1111.0 | No |
| Speech-02-HD · MiniMax | MiniMax | $100 /1M characters | 1109.0 | No |
| Turbo v2.5 · ElevenLabs | ElevenLabs | $50.0 /1M characters | 1099.0 | No |
| Multilingual v2 · ElevenLabs | ElevenLabs | $100 /1M characters | 1098.0 | No |
| TTS-1 HD · OpenAI | OpenAI | $30.0 /1M characters | 1097.0 | No |
| Chatterbox HD · Resemble AI | Resemble AI | $40.0 /1M characters | 1086.0 | No |
| TTS-1 · OpenAI | OpenAI | $15.0 /1M characters | 1084.0 | No |
| Speech-02-Turbo · MiniMax | MiniMax | $60.0 /1M characters | 1083.0 | No |
| Gemini 2.5 Flash Lite TTS · Google | $9.15 /1M characters | 1080.0 | No | |
| Flash v2.5 · ElevenLabs | ElevenLabs | $50.0 /1M characters | 1079.0 | No |
| Studio · Google | $160 /1M characters | 1073.0 | No | |
| Sonic 3 · Cartesia | Cartesia | $49.0 /1M characters | 1069.0 | No |
| Voxtral TTS · Mistral | Mistral | $16.0 /1M characters | 1067.0 | No |
| OpenAudio S1 · Fish Audio | Fish Audio | $15.0 /1M characters | 1066.0 | No |
| SIMBA 1.6 · SpeechifyAI | SpeechifyAI | $10.0 /1M characters | 1065.0 | No |
| Polly Generative · Amazon | Amazon | $30.0 /1M characters | 1064.0 | No |
| GPT-Realtime-2 · OpenAI | OpenAI | $192 /1M characters | 1062.0 | No |
| T2A-01-HD · MiniMax | MiniMax | $50.0 /1M characters | 1062.0 | No |
| Journey · Google | $160 /1M characters | 1056.0 | No | |
| Kokoro 82M v1.0 · Kokoro | Kokoro | $0.65 /1M characters | 1056.0 | No |
| Octave 2 · Hume AI | Hume AI | $87.5 /1M characters | 1052.0 | No |
| Chirp 3: HD · Google | $30.0 /1M characters | 1050.0 | No | |
| Async Flash v1.0 · async | async | $10.1 /1M characters | 1048.0 | No |
| Coda · Rime | Rime | $50.0 /1M characters | 1048.0 | No |
| OpenAudio S1 Mini · Fish Audio | Fish Audio | $15.0 /1M characters | 1045.0 | No |
| Gemini 2.5 Flash TTS (Dec 2025) · Google | $9.15 /1M characters | 1042.0 | No | |
| Polly Long-Form · Amazon | Amazon | $100 /1M characters | 1042.0 | No |
| Sonic English (Oct 2024) · Cartesia | Cartesia | $49.0 /1M characters | 1036.0 | No |
| SIMBA 1.0 · SpeechifyAI | SpeechifyAI | $10.0 /1M characters | 1034.0 | No |
| Gemini 2.5 Pro (Dec 2025) · Google | $18.3 /1M characters | 1032.0 | No | |
| Azure Neural · Microsoft | Microsoft | $15.0 /1M characters | 1026.0 | No |
| T2A-01-Turbo · MiniMax | MiniMax | $30.0 /1M characters | 1026.0 | No |
| Lightning v3.1 · Smallest.ai | Smallest.ai | $25.0 /1M characters | 1022.0 | No |
| Octave TTS · Hume AI | Hume AI | $87.5 /1M characters | 1021.0 | No |
| Chatterbox · Resemble AI | Resemble AI | $25.0 /1M characters | 1014.0 | No |
| Arcana v3 · Rime | Rime | $50.0 /1M characters | 1004.0 | No |
| Zonos-v0.1 · Zyphra | Zyphra | $20.0 /1M characters | 1000.0 | No |
| Murf Speech Gen 2 · Murf AI | Murf AI | $100 /1M characters | 975.0 | No |
| LMNT · LMNT | LMNT | $49.0 /1M characters | 970.0 | No |
| OpenVoice v2 · OpenVoice | OpenVoice | $8.33 /1M characters | 948.0 | No |
| Qwen3 TTS Flash · Alibaba | Alibaba | $10.0 /1M characters | 935.0 | No |
| Neuphonic TTS · Neuphonic | Neuphonic | $20.8 /1M characters | 932.0 | No |
| XTTS v2 · Coqui | Coqui | $40.4 /1M characters | 914.0 | No |
| WaveNet · Google | $16.0 /1M characters | 903.0 | No | |
| Mist V2 · Rime | Rime | $31.1 /1M characters | 891.0 | No |
| StyleTTS 2 · StyleTTS | StyleTTS | $2.82 /1M characters | 888.0 | No |
| Neural2 · Google | $16.0 /1M characters | 882.0 | No | |
| Polly Neural · Amazon | Amazon | $16.0 /1M characters | 880.0 | No |
| Standard · Google | $4.00 /1M characters | 874.0 | No | |
| Polly Standard · Amazon | Amazon | $4.00 /1M characters | 810.0 | No |
Compare models using the controlled-voice Arena and a shared eight-voice evaluation set.
Top 12 · Current models
Current voice cloning leaders using the source-native Arena Elo.
| Model | Creator | Arena Elo | 95% CI | Samples | Highlighted |
|---|---|---|---|---|---|
| Sonic 3.5 | Cartesia | 1105.2 | ±16.0 | 2,276 | No |
| Eleven v3 | ElevenLabs | 1063.2 | ±15.0 | 2,172 | No |
| Realtime TTS 1.5 Max | Inworld | 1059.5 | ±17.0 | 1,584 | No |
| StepAudio 2.5 TTS | StepFun | 1059.3 | ±16.0 | 1,537 | No |
| Realtime TTS-2 - Research Preview | Inworld | 1051.9 | ±16.0 | 1,558 | No |
| SpaceXAI TTS | SpaceXAI | 1047.4 | ±16.0 | 1,626 | No |
| Speech 2.8 HD | MiniMax | 1043.1 | ±15.0 | 2,173 | No |
| Simba 3.0 | SpeechifyAI | 1029.7 | ±16.0 | 1,597 | No |
| Speech 2.8 Turbo | MiniMax | 1028.0 | ±15.0 | 2,181 | No |
| Fish Audio S2 Pro | Fish Audio | 1023.5 | ±15.0 | 2,438 | No |
| Voxtral TTS | Mistral | 1019.2 | ±16.0 | 1,628 | No |
| Soniox TTS Real-Time v1 | Soniox | 1014.7 | ±15.0 | 2,006 | No |
USD per 1M characters
Price is decision context and never affects the source ranking.
| Model · Provider | Creator | Cost | Arena Elo | Highlighted |
|---|---|---|---|---|
| Sonic 3.5 · Cartesia | Cartesia | $49.0 /1M characters | 1105.2 | No |
| Eleven v3 · ElevenLabs | ElevenLabs | $100 /1M characters | 1063.2 | No |
| Realtime TTS 1.5 Max · Inworld | Inworld | $26.0 /1M characters | 1059.5 | No |
| StepAudio 2.5 TTS · StepFun | StepFun | $85.0 /1M characters | 1059.3 | No |
| Realtime TTS-2 - Research Preview · Inworld | Inworld | $20.8 /1M characters | 1051.9 | No |
| SpaceXAI TTS · SpaceXAI | SpaceXAI | $15.0 /1M characters | 1047.4 | No |
| Speech 2.8 HD · MiniMax | MiniMax | $100 /1M characters | 1043.1 | No |
| Simba 3.0 · SpeechifyAI | SpeechifyAI | $12.5 /1M characters | 1029.7 | No |
| Speech 2.8 Turbo · MiniMax | MiniMax | $60.0 /1M characters | 1028.0 | No |
| Fish Audio S2 Pro · Fish Audio | Fish Audio | $15.0 /1M characters | 1023.5 | No |
| Voxtral TTS · Mistral | Mistral | $16.0 /1M characters | 1019.2 | No |
| Soniox TTS Real-Time v1 · Soniox | Soniox | $14.2 /1M characters | 1014.7 | No |
| Multilingual v2 · ElevenLabs | ElevenLabs | $100 /1M characters | 1013.3 | No |
| Fish Audio S2.1 Pro · Fish Audio | Fish Audio | $15.0 /1M characters | 1004.7 | No |
| OpenAudio S1 Mini · Fish Audio | Fish Audio | $15.0 /1M characters | 1002.5 | No |
| OpenAudio S1 · Fish Audio | Fish Audio | $15.0 /1M characters | 1000.0 | No |
| Realtime TTS 1.5 Mini · Inworld | Inworld | $10.4 /1M characters | 998.5 | No |
| Fish Audio S1 · Fish Audio | Fish Audio | $15.0 /1M characters | 995.3 | No |
| Turbo v2.5 · ElevenLabs | ElevenLabs | $50.0 /1M characters | 975.5 | No |
| Flash v2.5 · ElevenLabs | ElevenLabs | $50.0 /1M characters | 974.8 | No |
| Lightning V3.1 Pro TTS (Jun 2026) · Smallest.ai | Smallest.ai | $19.5 /1M characters | 972.7 | No |
| Sonic 3 · Cartesia | Cartesia | $49.0 /1M characters | 939.4 | No |
| Chatterbox · Resemble AI | Resemble AI | $25.0 /1M characters | 926.5 | No |
| LMNT · LMNT | LMNT | $49.0 /1M characters | 912.7 | No |
| Lightning v3.1 · Smallest.ai | Smallest.ai | $25.0 /1M characters | 847.8 | No |
| XTTS v2 · Coqui | Coqui | $40.4 /1M characters | 815.0 | No |
| OpenVoice v2 · OpenVoice | OpenVoice | $8.33 /1M characters | 796.7 | No |
Compare the published Speech-to-Speech Index, latency, cost, and benchmark components.
Published index (%)
The published source index is retained exactly and is not a Lumina Score.
| Provider · Model | Index | Highlighted |
|---|---|---|
| Alibaba Cloud · Qwen Audio 3.0 Realtime Plus | 84.1% | No |
| SpaceXAI · Grok Voice Think Fast 2.0 High | 82.9% | No |
| OpenAI · GPT-Realtime-2.1 High | 79.1% | No |
| OpenAI · GPT-Realtime-2 (High) | 77.2% | No |
| Alibaba Cloud · Qwen Audio 3.0 Realtime Flash | 76.3% | No |
| SpaceXAI · Grok Voice Think Fast 1.0 | 75.7% | No |
| OpenAI · GPT-Realtime-2 (Medium) | 75.3% | No |
| OpenAI · GPT-Realtime-2.1 Minimal | 72.5% | No |
| OpenAI · GPT-Realtime-1.5 | 72.0% | No |
| Google · Gemini 3.1 Flash - High | 69.5% | No |
| OpenAI · GPT Realtime | 69.2% | No |
| OpenAI · GPT-Realtime-2 (Minimal) | 66.2% | No |
| OpenAI · GPT-Realtime-2.1 Mini High | 65.3% | No |
| SpaceXAI · Grok Voice Fast 1.0 | 64.1% | No |
| Deepslate · Deepslate Opal | 62.8% | No |
| OpenAI · GPT-Realtime-2.1 Mini Minimal | 59.0% | No |
| OpenAI · GPT Realtime Mini (Oct 2025) | 58.1% | No |
| Google · Gemini 3.1 Flash - Minimal | 56.6% | No |
Source-native component scores
Component measures remain separate and missing values remain unavailable.
| Provider · Model | BBA | FDB | τ-Voice | Highlighted |
|---|---|---|---|---|
| Alibaba Cloud · Qwen Audio 3.0 Realtime Plus | 99.0% | 98.4% | 54.6% | No |
| SpaceXAI · Grok Voice Think Fast 2.0 High | 97.2% | 95.1% | 56.5% | No |
| OpenAI · GPT-Realtime-2.1 High | 96.0% | 95.7% | 45.7% | No |
| OpenAI · GPT-Realtime-2 (High) | 97.0% | 95.3% | 39.8% | No |
| Alibaba Cloud · Qwen Audio 3.0 Realtime Flash | 96.0% | 96.9% | 35.9% | No |
| SpaceXAI · Grok Voice Think Fast 1.0 | 97.0% | 77.8% | 52.1% | No |
| OpenAI · GPT-Realtime-2 (Medium) | 93.0% | 95.2% | 37.4% | No |
| OpenAI · GPT-Realtime-2.1 Minimal | 87.0% | 92.7% | 38.0% | No |
| OpenAI · GPT-Realtime-1.5 | 81.0% | 95.7% | 38.8% | No |
| Google · Gemini 3.1 Flash - High | 97.0% | 74.3% | 37.7% | No |
Seconds to first audio
Latency is decision context and never affects the published index.
| Provider · Model | Index | Time to first audio | Highlighted |
|---|---|---|---|
| Alibaba Cloud · Qwen Audio 3.0 Realtime Plus | 84.1% | 4.02s | No |
| SpaceXAI · Grok Voice Think Fast 2.0 High | 82.9% | 0.70s | No |
| OpenAI · GPT-Realtime-2.1 High | 79.1% | 1.21s | No |
| OpenAI · GPT-Realtime-2 (High) | 77.2% | 1.14s | No |
| Alibaba Cloud · Qwen Audio 3.0 Realtime Flash | 76.3% | 4.16s | No |
| SpaceXAI · Grok Voice Think Fast 1.0 | 75.7% | 1.25s | No |
| OpenAI · GPT-Realtime-2 (Medium) | 75.3% | 1.22s | No |
| OpenAI · GPT-Realtime-2.1 Minimal | 72.5% | 0.97s | No |
| OpenAI · GPT-Realtime-1.5 | 72.0% | 0.81s | No |
| Google · Gemini 3.1 Flash - High | 69.5% | 2.99s | No |
| OpenAI · GPT Realtime | 69.2% | 0.98s | No |
| OpenAI · GPT-Realtime-2 (Minimal) | 66.2% | 1.12s | No |
| OpenAI · GPT-Realtime-2.1 Mini High | 65.3% | 4.28s | No |
| SpaceXAI · Grok Voice Fast 1.0 | 64.1% | 0.78s | No |
| Deepslate · Deepslate Opal | 62.8% | 0.44s | No |
| OpenAI · GPT-Realtime-2.1 Mini Minimal | 59.0% | 0.85s | No |
| OpenAI · GPT Realtime Mini (Oct 2025) | 58.1% | 0.81s | No |
| Google · Gemini 3.1 Flash - Minimal | 56.6% | 0.96s | No |
USD per hour of input audio
Input-hour cost is decision context and never affects the published index.
| Provider · Model | Index | Cost per hour | Highlighted |
|---|---|---|---|
| Alibaba Cloud · Qwen Audio 3.0 Realtime Plus | 84.1% | $4.42 | No |
| SpaceXAI · Grok Voice Think Fast 2.0 High | 82.9% | $4.80 | No |
| OpenAI · GPT-Realtime-2.1 High | 79.1% | $10.75 | No |
| OpenAI · GPT-Realtime-2 (High) | 77.2% | $4.14 | No |
| Alibaba Cloud · Qwen Audio 3.0 Realtime Flash | 76.3% | $4.77 | No |
| SpaceXAI · Grok Voice Think Fast 1.0 | 75.7% | $3.00 | No |
| OpenAI · GPT-Realtime-2 (Medium) | 75.3% | $3.97 | No |
| OpenAI · GPT-Realtime-2.1 Minimal | 72.5% | $11.31 | No |
| OpenAI · GPT-Realtime-1.5 | 72.0% | $11.44 | No |
| Google · Gemini 3.1 Flash - High | 69.5% | $1.75 | No |
| OpenAI · GPT Realtime | 69.2% | $11.08 | No |
| OpenAI · GPT-Realtime-2 (Minimal) | 66.2% | $3.07 | No |
| OpenAI · GPT-Realtime-2.1 Mini High | 65.3% | $3.45 | No |
| SpaceXAI · Grok Voice Fast 1.0 | 64.1% | $3.00 | No |
| Deepslate · Deepslate Opal | 62.8% | $6.48 | No |
| OpenAI · GPT-Realtime-2.1 Mini Minimal | 59.0% | $4.60 | No |
| OpenAI · GPT Realtime Mini (Oct 2025) | 58.1% | $3.04 | No |
| Google · Gemini 3.1 Flash - Minimal | 56.6% | $1.50 | No |
