Long text
How much text can it take in one request and still read every word?
Ranking
Score out of 100, higher is better. 5 items per model; the thin line on each bar is the 95% range.
When two models' ranges overlap, this test cannot tell them apart.
Longest clean read
The largest single request each model read with at least 97% of words correct. Longer is better.
Length by length
Word accuracy at each request size; darker is better. "Rejected" means the model refused that much text in one request. Hover a cell for details.
| Model | Score | 1,000 chars | 2,000 chars | 4,000 chars | 8,000 chars | 15,000 chars | |
|---|---|---|---|---|---|---|---|
| 1 | AnyToSpeech Premium and Charactersclean up to 15,000 | 100 | 100 | 100 | 100 | 100 | 100 |
| 2 | Kokoro 82Mclean up to 15,000 | 100 | 100 | 100 | 100 | 100 | 100 |
| 3 | Eleven Flash v2.5clean up to 15,000 | 100 | 100 | 100 | 100 | 99.9 | 100 |
| 4 | Gemini 3.8 Flash-Lite TTSclean up to 15,000 | 99.9 | 100 | 100 | 99.9 | 99.9 | 99.8 |
| 5 | AnyToSpeech Standard voicesclean up to 15,000 | 99.9 | 100 | 100 | 99.6 | 99.9 | 99.8 |
| 6 | Gemini 3.8 Flash TTSclean up to 15,000 | 99.9 | 100 | 99.5 | 100 | 99.9 | 99.9 |
| 7 | Grok TTSclean up to 15,000 | 99.8 | 100 | 99.7 | 99.6 | 99.8 | 99.9 |
| 8 | MiniMax Speech 2.8 HDclean up to 8,000 | 79.9 | 100 | 100 | 99.7 | 100 | rejected |
| 9 | MiniMax Speech 2.8 Turboclean up to 8,000 | 79.9 | 100 | 99.7 | 99.7 | 99.9 | rejected |
| 10 | Eleven Multilingual v2clean up to 8,000 | 79.9 | 100 | 100 | 99.7 | 99.7 | rejected |
| 11 | GPT-4o mini TTSclean up to 8,000 | 79.8 | 100 | 99.7 | 99.6 | 99.7 | rejected |
| 12 | Eleven v3clean up to 4,000 | 60 | 100 | 100 | 100 | rejected | rejected |
| 13 | TTS-1 HDclean up to 4,000 | 60 | 100 | 100 | 100 | rejected | rejected |
| 14 | Gemini 2.5 Pro TTSclean up to 4,000 | 59.9 | 100 | 100 | 99.7 | rejected | rejected |
| 15 | TTS-1clean up to 4,000 | 59.9 | 100 | 100 | 99.6 | rejected | rejected |
| 16 | Qwen3 TTSclean up to 2,000 | 58.6 | 100 | 99.5 | 62.1 | 29.5 | 1.9 |
| 17 | Inworld TTS-2clean up to 2,000 | 40 | 100 | 100 | rejected | rejected | rejected |
| 18 | Inworld TTS-1.5 Maxclean up to 2,000 | 40 | 100 | 100 | rejected | rejected | rejected |
| 19 | Chatterbox Multilingualclean up to nothing | 13.3 | 34.3 | 17.2 | 8.6 | 4.1 | 2.1 |
Products built on these models, AnyToSpeech included, split long documents into pieces before sending them. This test shows what the model itself does with one long request.
What we test and how it is scored
Test. One 15,000-character document, sent at five lengths: about 1,000, 2,000, 4,000, 8,000 and 15,000 characters. Each length is a single request with no splitting on our side, so every model is pushed up to and past its own limit.
Scoring. Each recording is transcribed in one-minute pieces and compared word for word with the text sent. A length scores its word accuracy; a length the model rejects, cuts short or garbles scores what comes back, which is 0 when nothing usable does. Score = average across the five lengths. "Longest clean read" is the largest length read with at least 97% of words correct.
Full item list and rules: methodology.
Think a score is wrong? Check it.
Every test item, scoring rule and known limit is published, and the audio behind each number is on the page.