OpenAIAudio modelDeprecated
GPT-4o Mini TTS
PopularityRank 354 of 876
36out of 100same as last month
Summary
GPT-4o Mini TTS is a speech-synthesis model built on GPT-4o Mini. It converts text into natural-sounding speech, accepts up to 2,000 input tokens, and produces audio only. Voice and delivery are controllable TTS outputs; a voice dialogue that understands and answers audio needs a separate conversation model.
On October 1, 2026, OpenAI announced that both listed TTS snapshots would shut down on January 6, 2027. The provider recommends GPT-Realtime 2.1 Mini as the replacement and points to the Realtime API for migration.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 87 published benchmark results
Page 1 of 8
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
Coval rolling-30-day pooled output word error rate (gpt-4o-mini-tts endpoint) (Show measurement, test conditions, and source)
- Source value
- 1.8020686376142856
- Score
- 1.8020686376142856
- Metric
- Pooled word error rate of speech synthesized from Coval's fixed text prompts
- Unit
- %
- Benchmark version
- Coval rolling 30-day snapshot, last measured 2026-10-16, source response fetched 2026-10-16T15:36:03.979138Z
- Category
- audio
- Direction
- Lower is better
- Sample
- 2012
- Source type
- independent-evaluator
- Evaluator
- Coval
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Independent evaluation. Exact hosted API model ID gpt-4o-mini-tts. Coval describes a fixed voice and prompt configuration, daily measurements, and a rolling 30-day result last measured 2026-10-16. Stored score is pooled_value=1.8020686376142856 from the WER row, not mean_value; Coval's visible 1.8% is rounded to one decimal. The reported WER sample count is from the same row.
- Context
- This is a transcript-based intelligibility check on generated speech, not a human preference or naturalness rating. The page publishes WER-scored samples, not total attempted requests or a failure count, and gives no confidence interval. The fixed voice and instruction do not represent the model's full expressive range.
EmergentTTS-Eval unprompted WER, Emotions (Show measurement, test conditions, and source)
- Source value
- 0.72
- Score
- 0.72
- Metric
- Word error rate (WER)
- Unit
- %
- Benchmark version
- NeurIPS 2025 Datasets and Benchmarks Track, Table 1
- Category
- audio
- Direction
- Lower is better
- Sample
- 280
- Source type
- independent-evaluator
- Evaluator
- EmergentTTS-Eval, NeurIPS 2025
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Published API model label gpt-4o-mini-tts, Alloy voice, unprompted baseline using the utterance text. WER was computed with Whisper-v3-large. The source reports 280 samples for this Emotions row.
- Context
- The paper names the API alias but publishes no immutable serving revision or confidence interval. The result concerns transcription-based intelligibility for the benchmark utterances, not general TTS quality.
EmergentTTS-Eval strong-prompt WER, Emotions (Show measurement, test conditions, and source)
- Source value
- 0.71
- Score
- 0.71
- Metric
- Word error rate (WER)
- Unit
- %
- Benchmark version
- NeurIPS 2025 Datasets and Benchmarks Track, Table 1
- Category
- audio
- Direction
- Lower is better
- Sample
- 280
- Source type
- independent-evaluator
- Evaluator
- EmergentTTS-Eval, NeurIPS 2025
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Published API model label gpt-4o-mini-tts, Alloy voice, category-specific strong prompt. The paper describes category instructions added to the utterance. WER was computed with Whisper-v3-large. The source reports 280 samples for this Emotions row.
- Context
- This is the same API alias under a separate prompt configuration, not a separate model checkpoint. No immutable serving revision or confidence interval is published.
EmergentTTS-Eval strong-prompt win rate vs baseline, Emotions, Gemini 2.5 Pro Preview (Show measurement, test conditions, and source)
- Source value
- 59.17
- Score
- 59.17
- Metric
- Model-as-a-judge preference win rate versus unprompted gpt-4o-mini-tts Alloy baseline
- Unit
- %
- Benchmark version
- NeurIPS 2025 Table 1; original leaderboard judge gemini-2.5-pro-preview-05-06
- Category
- audio
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- EmergentTTS-Eval, NeurIPS 2025
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Published API model label gpt-4o-mini-tts, Alloy voice, strong prompt compared with its unprompted gpt-4o-mini-tts Alloy baseline. The original paper leaderboard used gemini-2.5-pro-preview-05-06 as judge at temperature 0.0. The row is a judged preference rate; the source does not publish the row denominator or a confidence interval.
- Context
- The judge version is distinct from the later non-preview leaderboard. These are separately configured evaluations, not conflicting claims for an identical judging run. The API serving revision is not pinned.
EmergentTTS-Eval strong-prompt win rate vs baseline, Emotions, Gemini 2.5 Pro (Show measurement, test conditions, and source)
- Source value
- 67.32
- Score
- 67.32
- Metric
- Model-as-a-judge preference win rate versus unprompted gpt-4o-mini-tts Alloy baseline
- Unit
- %
- Benchmark version
- Pinned EmergentTTS-Eval README leaderboard; gemini-2.5-pro, thinking budget 256; repository commit fe02c5d27f2d87751a66e32631aaa73a92d9cbbb
- Category
- audio
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- EmergentTTS-Eval pinned README leaderboard
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Published API model label gpt-4o-mini-tts, Alloy voice, strong prompt compared with the unprompted gpt-4o-mini-tts Alloy baseline. The pinned README identifies the current leaderboard judge as non-preview gemini-2.5-pro with thinking budget 256. The row is a judged preference rate; the source does not publish the row denominator or a confidence interval.
- Context
- The pinned README distinguishes this non-preview judge run from the original preview-judge leaderboard. Evaluation date and immutable OpenAI serving revision are not published.
EmergentTTS-Eval unprompted WER, Foreign Words (Show measurement, test conditions, and source)
- Source value
- 13.45
- Score
- 13.45
- Metric
- Word error rate (WER)
- Unit
- %
- Benchmark version
- NeurIPS 2025 Datasets and Benchmarks Track, Table 1
- Category
- audio
- Direction
- Lower is better
- Sample
- 280
- Source type
- independent-evaluator
- Evaluator
- EmergentTTS-Eval, NeurIPS 2025
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Published API model label gpt-4o-mini-tts, Alloy voice, unprompted baseline using the utterance text. WER was computed with Whisper-v3-large. The source reports 280 samples for this Foreign Words row.
- Context
- The paper names the API alias but publishes no immutable serving revision or confidence interval. The result concerns transcription-based intelligibility for the benchmark utterances, not general TTS quality.
EmergentTTS-Eval strong-prompt WER, Foreign Words (Show measurement, test conditions, and source)
- Source value
- 12.07
- Score
- 12.07
- Metric
- Word error rate (WER)
- Unit
- %
- Benchmark version
- NeurIPS 2025 Datasets and Benchmarks Track, Table 1
- Category
- audio
- Direction
- Lower is better
- Sample
- 280
- Source type
- independent-evaluator
- Evaluator
- EmergentTTS-Eval, NeurIPS 2025
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Published API model label gpt-4o-mini-tts, Alloy voice, category-specific strong prompt. The paper describes category instructions added to the utterance. WER was computed with Whisper-v3-large. The source reports 280 samples for this Foreign Words row.
- Context
- This is the same API alias under a separate prompt configuration, not a separate model checkpoint. No immutable serving revision or confidence interval is published.
EmergentTTS-Eval strong-prompt win rate vs baseline, Foreign Words, Gemini 2.5 Pro Preview (Show measurement, test conditions, and source)
- Source value
- 57.32
- Score
- 57.32
- Metric
- Model-as-a-judge preference win rate versus unprompted gpt-4o-mini-tts Alloy baseline
- Unit
- %
- Benchmark version
- NeurIPS 2025 Table 1; original leaderboard judge gemini-2.5-pro-preview-05-06
- Category
- audio
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- EmergentTTS-Eval, NeurIPS 2025
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Published API model label gpt-4o-mini-tts, Alloy voice, strong prompt compared with its unprompted gpt-4o-mini-tts Alloy baseline. The original paper leaderboard used gemini-2.5-pro-preview-05-06 as judge at temperature 0.0. The row is a judged preference rate; the source does not publish the row denominator or a confidence interval.
- Context
- The judge version is distinct from the later non-preview leaderboard. These are separately configured evaluations, not conflicting claims for an identical judging run. The API serving revision is not pinned.
EmergentTTS-Eval strong-prompt win rate vs baseline, Foreign Words, Gemini 2.5 Pro (Show measurement, test conditions, and source)
- Source value
- 63.03
- Score
- 63.03
- Metric
- Model-as-a-judge preference win rate versus unprompted gpt-4o-mini-tts Alloy baseline
- Unit
- %
- Benchmark version
- Pinned EmergentTTS-Eval README leaderboard; gemini-2.5-pro, thinking budget 256; repository commit fe02c5d27f2d87751a66e32631aaa73a92d9cbbb
- Category
- audio
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- EmergentTTS-Eval pinned README leaderboard
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Published API model label gpt-4o-mini-tts, Alloy voice, strong prompt compared with the unprompted gpt-4o-mini-tts Alloy baseline. The pinned README identifies the current leaderboard judge as non-preview gemini-2.5-pro with thinking budget 256. The row is a judged preference rate; the source does not publish the row denominator or a confidence interval.
- Context
- The pinned README distinguishes this non-preview judge run from the original preview-judge leaderboard. Evaluation date and immutable OpenAI serving revision are not published.
EmergentTTS-Eval unprompted WER, Paralinguistics (Show measurement, test conditions, and source)
- Source value
- 20.55
- Score
- 20.55
- Metric
- Word error rate (WER)
- Unit
- %
- Benchmark version
- NeurIPS 2025 Datasets and Benchmarks Track, Table 1
- Category
- audio
- Direction
- Lower is better
- Sample
- 280
- Source type
- independent-evaluator
- Evaluator
- EmergentTTS-Eval, NeurIPS 2025
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Published API model label gpt-4o-mini-tts, Alloy voice, unprompted baseline using the utterance text. WER was computed with Whisper-v3-large. The source reports 280 samples for this Paralinguistics row.
- Context
- The paper names the API alias but publishes no immutable serving revision or confidence interval. The result concerns transcription-based intelligibility for the benchmark utterances, not general TTS quality.
EmergentTTS-Eval strong-prompt WER, Paralinguistics (Show measurement, test conditions, and source)
- Source value
- 21.33
- Score
- 21.33
- Metric
- Word error rate (WER)
- Unit
- %
- Benchmark version
- NeurIPS 2025 Datasets and Benchmarks Track, Table 1
- Category
- audio
- Direction
- Lower is better
- Sample
- 280
- Source type
- independent-evaluator
- Evaluator
- EmergentTTS-Eval, NeurIPS 2025
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Published API model label gpt-4o-mini-tts, Alloy voice, category-specific strong prompt. The paper describes category instructions added to the utterance. WER was computed with Whisper-v3-large. The source reports 280 samples for this Paralinguistics row.
- Context
- This is the same API alias under a separate prompt configuration, not a separate model checkpoint. No immutable serving revision or confidence interval is published.
EmergentTTS-Eval strong-prompt win rate vs baseline, Paralinguistics, Gemini 2.5 Pro Preview (Show measurement, test conditions, and source)
- Source value
- 58.75
- Score
- 58.75
- Metric
- Model-as-a-judge preference win rate versus unprompted gpt-4o-mini-tts Alloy baseline
- Unit
- %
- Benchmark version
- NeurIPS 2025 Table 1; original leaderboard judge gemini-2.5-pro-preview-05-06
- Category
- audio
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- EmergentTTS-Eval, NeurIPS 2025
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Published API model label gpt-4o-mini-tts, Alloy voice, strong prompt compared with its unprompted gpt-4o-mini-tts Alloy baseline. The original paper leaderboard used gemini-2.5-pro-preview-05-06 as judge at temperature 0.0. The row is a judged preference rate; the source does not publish the row denominator or a confidence interval.
- Context
- The judge version is distinct from the later non-preview leaderboard. These are separately configured evaluations, not conflicting claims for an identical judging run. The API serving revision is not pinned.
Page 1 of 8
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
0 to 100, higher is more popular
20 of 83 models
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 6, 2026
GPT-4o Mini TTSFlux 37Gemini 3.5 Transcribe Live 35GPT-Realtime Translate 35GPT-Realtime-Whisper 34
Method changes in the timeline, a jump at these points is no market event:
- 1November 1, 2026: Value only with search or usage data, 30-day attention
What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.
GPT-4o Mini TTS compared
- 1GPT-4o Mini TTS vs. Sonic 3.66 shared tests
- 2GPT-4o Mini TTS vs. Eleven v39 shared tests
- 3GPT-4o Mini TTS vs. Fish Audio S2 Pro1 shared test
- 4GPT-4o Mini TTS vs. Gemini 3.1 Flash TTS9 shared tests
- 5GPT-4o Mini TTS vs. Voxtral Mini TTS1 shared test
- 6GPT-4o Mini TTS vs. Eleven v49 shared tests
- 7GPT-4o Mini TTS vs. Eleven v4 Turbo2 shared tests
- 8GPT-4o Mini TTS vs. Fish Audio S2.1 Pro3 shared tests
- 9GPT-4o Mini TTS vs. Gemini 3.8 Flash-Lite TTS7 shared tests
Similar models
More models of the same kind as GPT-4o Mini TTS, chosen by model line, developer, size, price and release date.
- GPT-Realtime 2.1OpenAI · same developer
- GPT-Live-TranscribeOpenAI · same developer
- GPT-Realtime TranslateOpenAI · same developer
- GPT-Realtime-WhisperOpenAI · same developer
- GPT-TranscribeOpenAI · same developer
- GPT-Realtime 2.1 MiniOpenAI · same developer
- GPT-Live 1OpenAI · same developer
- Gemini 3.5 TranscribeGoogle · popular in this category
All specifications
Architecture
Input and output
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
API
Further details
- Shutdown
- Both listed snapshots shut down on January 6, 2027; OpenAI recommends GPT-Realtime 2.1 Mini
- Source: Shutdown
Sources and data date
Show all 40 sources
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.
Without vendor claims and historical or disputed results; those stay labelled in the single values.
| Type | Evidence and data date |
|---|---|
| Research status | Research date October 15, 2026. 34 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not published | OpenAI API serving revision and EmergentTTS judge denominators Note The EmergentTTS sources publish the gpt-4o-mini-tts alias and judge identity but no immutable OpenAI serving revision, per-row judge denominator, or confidence interval. |
| Not published | human-listening-study Note Audar explicitly says a formal human MOS/CMOS listening study is planned and has not yet been completed. |
| Unresolved | arabic-dialect-row-sample-size-and-evaluation-version Note The inspected Arabic dialect table does not state per-row sample sizes, evaluation date, model/API revision, or uncertainty. Its separate n=2,902 full-MSA tie-breaker must not be attributed to the table's GPT-4o-mini-TTS row. |
| Not found in the inspected sources | Current Coval endpoint benchmark values Note The exact model endpoint is no longer available on its public model page, and the live task leaderboards do not contain an exact row. Historic research is retained; this refresh has no numeric score. |
| Additional source | OpenAI model documentation (retrieved October 2, 2026; October 15, 2026; October 16, 2026) · OpenAI GPT-4o Mini TTS model documentation · Editorial description reviewed October 2, 2026 |
| Additional source | Model lifecycle (retrieved October 2, 2026) · OpenAI text-to-speech model deprecations · Editorial description reviewed October 2, 2026 |
| Additional source | SpeechArena (retrieved October 15, 2026) |
| Additional source | Coval (retrieved October 16, 2026; October 17, 2026) (unavailable on November 2, 2026) https://benchmarks.coval.ai/models/gpt-4o-mini-tts |
| Additional source | EmergentTTS-Eval, NeurIPS 2025 (retrieved October 17, 2026) |
| Additional source | EmergentTTS-Eval pinned README leaderboard (retrieved October 17, 2026) |
| Additional source | Rime Labs TTSBench SupportBench (retrieved October 17, 2026) |
| Additional source | Rime Labs TTSBench ContentBench (retrieved October 17, 2026) |
| Additional source | InstructTTSEval authors, InstructTTSEval: Benchmarking Complex Natural-Language Instruction Following in Text-to-Speech Systems (arXiv:2506.16381v1) (retrieved October 17, 2026) |
| Additional source | Audar-TTS-V1 benchmark authors (retrieved October 19, 2026) |
| Additional source | ACL Anthology paper 2026.acl-short.59 (retrieved October 19, 2026) |
| Additional source | Do You Hear What I Mean? Quantifying the Instruction-Perception Gap in Instruction-Guided Expressive Text-To-Speech Systems (retrieved October 20, 2026) |
| Additional source | Rime Labs TTSBench AlphaBench (retrieved October 20, 2026) |
| Additional source | Xue et al., NVV-SuperBench, arXiv:2604.16211v3, Table 4 (retrieved October 21, 2026) |
| Additional source | Coval Voice AI Benchmarks (retrieved October 23, 2026) |
| Additional source | DesignArena (retrieved October 28, 2026)12 |
| Additional source | Speko (retrieved October 28, 2026) |
| Additional source | VoiceArena primary public leaderboard (retrieved October 28, 2026) |
| Additional source | Model metadata (retrieved October 15, 2026) |
| Additional source | Model metadata (retrieved October 17, 2026)123456 |
| Additional source | Model metadata (retrieved October 20, 2026) |
| Additional source | Model metadata (retrieved October 28, 2026)1234567 |
| Additional source | OpenRouter: Popularity index (derived), CC BY 4.0 |
| Additional source | Vercel AI Gateway: Popularity index (derived), CC BY 4.0 |
| Additional source | Ollama: cumulative downloads for the popularity index |
| Additional source | Replicate: cumulative runs for the popularity index |
| Additional source | Google Trending Now: limited feed selection for Germany and the US |
| Additional source | Creative Commons Attribution 4.0 |