whatmodel.sh / tts / GPT-4o mini TTS
GPT-4o mini TTS
↗ share
+ contribute
GPT-4o mini TTS is a closed-weights text-to-speech model from OpenAI, released March 2025.
Catalogue identity notes
gpt-4o-mini-tts as explicitly linked in the result row.
Model information sources
Score within the Vapi Humanness · Bradley–Terry rating, 2026-09-25 snapshot comparison set only.
$ how it compares
The text-to-speech models ranked either side of GPT-4o mini TTS, out of 148 with a score, then more from OpenAI.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 56 |  Octave TTS (Octave 1) | 58±4 | api $50.00/M chars | ||
| 57 |  Sonic 3 | 58±2 | api $65.00/M chars | ||
| 58 | GPT-4o mini TTS | 58±2 | comparison only | ||
| 59 | Sonic 2 | 58±3 | comparison only | ||
| 60 |  F5-TTS Base (v0) | 57±2 | local | ||
| more from OpenAI | |||||
| 52 | TTS-1 | 59±4 | api $15.00/M chars | ||
| 61 | TTS-1 HD | 57±7 | api $30.00/M chars | ||
see all text-to-speech models · find the ones that fit your machine
$ what it's good at
other tests12 not in a task · 98% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| InstructTTSEval ZH · mean instruction-following score (APS, DSD, RP) | 51.1% ✎ | ▄··········█ beats 0/2 | 16% |
| EmergentTTS · 6-category preference | 60.2% ✎ | ▅▅▃▃█▅▅▃·▃·▅ beats 14/16 | 16% |
| RW-Voice-EQ TTS · Acoustic Quality (human) | 3.34 / 5 ✎ | █·█········▄ beats 2/4 | 13% |
| Rime ContentBench · paired human preference vs Rime Coda (US English) | -0.31 / 2 ✎ | ▄·▄········█ beats 0/3 | 10% |
| Rime SupportBench · paired human preference vs Rime Coda (US English) | -0.90 / 2 ✎ | █·······█·██ beats 0/3 | 9.5% |
| RW-Voice-EQ TTS · Language Code-Switching (human) | 3.34 / 5 ✎ | █··▄·······▄ beats 1/3 | 8.8% |
| Voice-First Nation · 10 Indian languages overall pairwise preference (Bradley–Terry) | 942.8 ✎ | ▄····▄·····█ beats 0/3 | 6.8% |
| InstructTTSEval EN · mean instruction-following score (APS, DSD, RP) | 68.5% ✎ | ▄····█··▄··▄ beats 1/4 | 6.6% |
4 more tests
| test | result | vs the others in its run | counts |
|---|---|---|---|
| RW-Voice-EQ TTS · Expressiveness (human) | 3.23 / 5 ✎ | ███···█····█ beats 0/4 | 4.1% |
| EmergentTTS · v1 paper automatic MOS (WV-MOS) | 4.20 / 5 ✎ | ▄·█·█▄····█▄ beats 7/8 | 3.7% |
| EmergentTTS · 6-category WER | 10.8% ✎ | ▂▄··▂▄▄··█▄▆ beats 14/16 | 3.5% |
| Amharic TTS · 87-prompt CER (Chirp 3 judge) | 18.7% ✎ | ▄·····▄·▄·██ beats 1/6 | 0.0% |
$ where to run it
No verified API or download recorded yet · add one
$ how the score was worked out · RW-Voice-EQ Bench (Hume AI) 26%, InstructTTSEval: instruction-… 22%, Rime Alpha TTS Benchmark v1.0… 20% and 4 more
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 RW-Voice-EQ Bench (Hume AI) | 3 | 26% |
| InstructTTSEval: instruction-following TTS benchmark (arXiv:2506.16381v1) | 2 | 22% |
 Rime Alpha TTS Benchmark v1.0.0 · SupportBench and ContentBench paired human preference | 2 | 20% |
 EmergentTTS-Eval six-category leaderboard | 2 | 19% |
| Preferences of a Voice-First Nation: pairwise human evaluation of TTS in 10 Indian languages (arXiv:2604.21481v2) | 1 | 7% |
| EmergentTTS-Eval v1 paper (arXiv:2505.23009v1) · Table 1 automatic MOS | 1 | 4% |
| Addis AI Amharic TTS benchmark | 1 | 0% |
| prior · pull toward the middle | 2% | |
| weighted average | = 57.6 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json