OpenAIAudio modelActive
GPT-Realtime Translate
PopularityRank 364 of 876
35out of 100same as last month
Summary
GPT-Realtime-Translate is a narrowly focused speech-to-speech translation model. It consumes incoming audio, streams translated audio, and returns transcript deltas while the source speech is still arriving. The model uses a dedicated Realtime translation endpoint. It is not meant for general reasoning or batch document translation.
GPT-Realtime-Translate uses a 16,000-token context and returns up to 2,000 tokens. OpenAI bills it by audio duration rather than text tokens. Gemini 3.5 Live Translate serves a similar speech-to-speech role and Google documents support for more than 70 languages; GPT-Realtime-Translate also returns transcript deltas.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 40 published benchmark results
Page 1 of 4
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.
LiveLingo 2026 comprehension fidelity composite (Show measurement, test conditions, and source)
- Source value
- 4.53
- Score
- 4.53
- Metric
- two-LLM-judge composite on a 0-5 scale
- Unit
- rating
- Benchmark version
- LiveLingo Research benchmark addendum, evaluated 2026-06-10
- Category
- audio
- Direction
- Higher is better
- Sample
- 120
- Source type
- vendor-reported
- Evaluator
- LiveLingo Research
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact gpt-realtime-translate API model; 120 utterances across four language pairs (English to Spanish, Chinese, Japanese, and German); GPT-4o and Gemini 2.5 Flash judges used the stated rubric and prompts. The publisher reports 4 wins, 80 ties, and 36 losses against LiveLingo at the utterance-cell level.
- Context
- LiveLingo is published by a company selling a competing translation product and explicitly discloses its financial interest. The benchmark is publisher-reported, not an independently run evaluation; the two judge models do not remove that conflict. No independent exact-model quality result was found. The page describes the four-language-pair quality cohort separately from its continuous-speech latency examples.
LiveLingo median time to first translated audio (Show measurement, test conditions, and source)
- Source value
- 0.711
- Score
- 0.711
- Metric
- median time from speech onset to first translated audio
- Unit
- seconds
- Benchmark version
- LiveLingo Research benchmark addendum, evaluated 2026-06-10
- Category
- audio
- Direction
- Lower is better
- Sample
- 120
- Source type
- vendor-reported
- Evaluator
- LiveLingo Research
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact gpt-realtime-translate API model; 120 evaluated sessions; source speech start to first translated audio; publisher reports p10-p90 of 485-1,012 ms.
- Context
- LiveLingo is a financially interested competitor, so this is a vendor-reported measurement. The public raw addendum contains 120 values whose middle pair is 708 and 711 ms, yielding a derived median of 709.5 ms; the page publishes 711 ms. This record retains only the publisher's stated 711 ms and does not present the recomputation as a second benchmark. No measurement protocol beyond the described start and endpoint is provided.
LiveLingo median utterance-end-to-translation-arrival latency (Show measurement, test conditions, and source)
- Source value
- 3.8
- Score
- 3.8
- Metric
- median delay from source utterance end to arrival of its translated speech
- Unit
- seconds
- Benchmark version
- LiveLingo Research benchmark addendum, evaluated 2026-06-10
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- LiveLingo Research
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. The publisher describes continuous speech on 120-second news clips and reports a median 3.8-second lag, increasing to 20.3 seconds on dense audio. It does not publish a sample count for this latency statistic; none is inferred.
- Context
- LiveLingo is a financially interested competitor and the result is vendor-reported. It is a separate latency metric from first-audio latency and from the 120-utterance comprehension score. The source does not provide the per-model sample count or enough detail to reproduce the statistic.
OpenSTBench BLEU, EN→ZH (Show measurement, test conditions, and source)
- Source value
- 21.4059
- Score
- 21.4059
- Metric
- BLEU score (points); higher is better
- Unit
- points
- Benchmark version
- OpenSTBench arXiv:2605.30792v1 (Table 5), 2026-05-29
- Direction
- Higher is better
- Sample
- 1000
- Source type
- independent-evaluator
- Evaluator
- OpenSTBench authors, OpenSTBench: Beyond Semantic Evaluation for Speech Translation (arXiv:2605.30792v1)
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. OpenSTBench arXiv:2605.30792v1, exact system label GPT Realtime Translate, evaluated as an online streaming speech-to-speech API system and linked by the paper to OpenAI's gpt-realtime-translate model documentation. The benchmark evaluates a system/API route, not a disclosed fixed weight checkpoint. Translation, speech-quality, temporal-consistency, and MSLT metrics use N=1,000 per direction; speaker-preservation metrics use a 300-example paired LibriTTS set; emotion metrics use RAVDESS EN→ZH N=1,440 and MCAE-SPPS ZH→EN N=1,029; nonverbal-event metrics use NonverbalTTS EN→ZH N=359 and SynParaSpeech ZH→EN N=500. The paper does not publish an immutable API serving revision or confidence intervals.
- Context
- Independent evaluator preprint reports a named API/system route. This is not an isolated-model or fixed-checkpoint result: OpenAI's serving revision, backend hardware, decoding controls, and API parameters are not pinned. No numeric confidence interval is reported. The source paper version is arXiv:2605.30792v1 (2026-05-29).
OpenSTBench chrF++, EN→ZH (Show measurement, test conditions, and source)
- Source value
- 14.2556
- Score
- 14.2556
- Metric
- chrF++ score (points); higher is better
- Unit
- points
- Benchmark version
- OpenSTBench arXiv:2605.30792v1 (Table 5), 2026-05-29
- Direction
- Higher is better
- Sample
- 1000
- Source type
- independent-evaluator
- Evaluator
- OpenSTBench authors, OpenSTBench: Beyond Semantic Evaluation for Speech Translation (arXiv:2605.30792v1)
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. OpenSTBench arXiv:2605.30792v1, exact system label GPT Realtime Translate, evaluated as an online streaming speech-to-speech API system and linked by the paper to OpenAI's gpt-realtime-translate model documentation. The benchmark evaluates a system/API route, not a disclosed fixed weight checkpoint. Translation, speech-quality, temporal-consistency, and MSLT metrics use N=1,000 per direction; speaker-preservation metrics use a 300-example paired LibriTTS set; emotion metrics use RAVDESS EN→ZH N=1,440 and MCAE-SPPS ZH→EN N=1,029; nonverbal-event metrics use NonverbalTTS EN→ZH N=359 and SynParaSpeech ZH→EN N=500. The paper does not publish an immutable API serving revision or confidence intervals.
- Context
- Independent evaluator preprint reports a named API/system route. This is not an isolated-model or fixed-checkpoint result: OpenAI's serving revision, backend hardware, decoding controls, and API parameters are not pinned. No numeric confidence interval is reported. The source paper version is arXiv:2605.30792v1 (2026-05-29).
OpenSTBench COMET, EN→ZH (Show measurement, test conditions, and source)
- Source value
- 0.787
- Score
- 0.787
- Metric
- COMET score; higher is better
- Unit
- ratio
- Benchmark version
- OpenSTBench arXiv:2605.30792v1 (Table 5), 2026-05-29
- Direction
- Higher is better
- Sample
- 1000
- Source type
- independent-evaluator
- Evaluator
- OpenSTBench authors, OpenSTBench: Beyond Semantic Evaluation for Speech Translation (arXiv:2605.30792v1)
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. OpenSTBench arXiv:2605.30792v1, exact system label GPT Realtime Translate, evaluated as an online streaming speech-to-speech API system and linked by the paper to OpenAI's gpt-realtime-translate model documentation. The benchmark evaluates a system/API route, not a disclosed fixed weight checkpoint. Translation, speech-quality, temporal-consistency, and MSLT metrics use N=1,000 per direction; speaker-preservation metrics use a 300-example paired LibriTTS set; emotion metrics use RAVDESS EN→ZH N=1,440 and MCAE-SPPS ZH→EN N=1,029; nonverbal-event metrics use NonverbalTTS EN→ZH N=359 and SynParaSpeech ZH→EN N=500. The paper does not publish an immutable API serving revision or confidence intervals.
- Context
- Independent evaluator preprint reports a named API/system route. This is not an isolated-model or fixed-checkpoint result: OpenAI's serving revision, backend hardware, decoding controls, and API parameters are not pinned. No numeric confidence interval is reported. The source paper version is arXiv:2605.30792v1 (2026-05-29).
OpenSTBench BLEURT, EN→ZH (Show measurement, test conditions, and source)
- Source value
- 0.5816
- Score
- 0.5816
- Metric
- BLEURT score; higher is better
- Unit
- ratio
- Benchmark version
- OpenSTBench arXiv:2605.30792v1 (Table 5), 2026-05-29
- Direction
- Higher is better
- Sample
- 1000
- Source type
- independent-evaluator
- Evaluator
- OpenSTBench authors, OpenSTBench: Beyond Semantic Evaluation for Speech Translation (arXiv:2605.30792v1)
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. OpenSTBench arXiv:2605.30792v1, exact system label GPT Realtime Translate, evaluated as an online streaming speech-to-speech API system and linked by the paper to OpenAI's gpt-realtime-translate model documentation. The benchmark evaluates a system/API route, not a disclosed fixed weight checkpoint. Translation, speech-quality, temporal-consistency, and MSLT metrics use N=1,000 per direction; speaker-preservation metrics use a 300-example paired LibriTTS set; emotion metrics use RAVDESS EN→ZH N=1,440 and MCAE-SPPS ZH→EN N=1,029; nonverbal-event metrics use NonverbalTTS EN→ZH N=359 and SynParaSpeech ZH→EN N=500. The paper does not publish an immutable API serving revision or confidence intervals.
- Context
- Independent evaluator preprint reports a named API/system route. This is not an isolated-model or fixed-checkpoint result: OpenAI's serving revision, backend hardware, decoding controls, and API parameters are not pinned. No numeric confidence interval is reported. The source paper version is arXiv:2605.30792v1 (2026-05-29).
OpenSTBench BLEU, ZH→EN (Show measurement, test conditions, and source)
- Source value
- 16.1659
- Score
- 16.1659
- Metric
- BLEU score (points); higher is better
- Unit
- points
- Benchmark version
- OpenSTBench arXiv:2605.30792v1 (Table 5), 2026-05-29
- Direction
- Higher is better
- Sample
- 1000
- Source type
- independent-evaluator
- Evaluator
- OpenSTBench authors, OpenSTBench: Beyond Semantic Evaluation for Speech Translation (arXiv:2605.30792v1)
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. OpenSTBench arXiv:2605.30792v1, exact system label GPT Realtime Translate, evaluated as an online streaming speech-to-speech API system and linked by the paper to OpenAI's gpt-realtime-translate model documentation. The benchmark evaluates a system/API route, not a disclosed fixed weight checkpoint. Translation, speech-quality, temporal-consistency, and MSLT metrics use N=1,000 per direction; speaker-preservation metrics use a 300-example paired LibriTTS set; emotion metrics use RAVDESS EN→ZH N=1,440 and MCAE-SPPS ZH→EN N=1,029; nonverbal-event metrics use NonverbalTTS EN→ZH N=359 and SynParaSpeech ZH→EN N=500. The paper does not publish an immutable API serving revision or confidence intervals.
- Context
- Independent evaluator preprint reports a named API/system route. This is not an isolated-model or fixed-checkpoint result: OpenAI's serving revision, backend hardware, decoding controls, and API parameters are not pinned. No numeric confidence interval is reported. The source paper version is arXiv:2605.30792v1 (2026-05-29).
OpenSTBench chrF++, ZH→EN (Show measurement, test conditions, and source)
- Source value
- 39.8521
- Score
- 39.8521
- Metric
- chrF++ score (points); higher is better
- Unit
- points
- Benchmark version
- OpenSTBench arXiv:2605.30792v1 (Table 5), 2026-05-29
- Direction
- Higher is better
- Sample
- 1000
- Source type
- independent-evaluator
- Evaluator
- OpenSTBench authors, OpenSTBench: Beyond Semantic Evaluation for Speech Translation (arXiv:2605.30792v1)
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. OpenSTBench arXiv:2605.30792v1, exact system label GPT Realtime Translate, evaluated as an online streaming speech-to-speech API system and linked by the paper to OpenAI's gpt-realtime-translate model documentation. The benchmark evaluates a system/API route, not a disclosed fixed weight checkpoint. Translation, speech-quality, temporal-consistency, and MSLT metrics use N=1,000 per direction; speaker-preservation metrics use a 300-example paired LibriTTS set; emotion metrics use RAVDESS EN→ZH N=1,440 and MCAE-SPPS ZH→EN N=1,029; nonverbal-event metrics use NonverbalTTS EN→ZH N=359 and SynParaSpeech ZH→EN N=500. The paper does not publish an immutable API serving revision or confidence intervals.
- Context
- Independent evaluator preprint reports a named API/system route. This is not an isolated-model or fixed-checkpoint result: OpenAI's serving revision, backend hardware, decoding controls, and API parameters are not pinned. No numeric confidence interval is reported. The source paper version is arXiv:2605.30792v1 (2026-05-29).
OpenSTBench COMET, ZH→EN (Show measurement, test conditions, and source)
- Source value
- 0.6606
- Score
- 0.6606
- Metric
- COMET score; higher is better
- Unit
- ratio
- Benchmark version
- OpenSTBench arXiv:2605.30792v1 (Table 5), 2026-05-29
- Direction
- Higher is better
- Sample
- 1000
- Source type
- independent-evaluator
- Evaluator
- OpenSTBench authors, OpenSTBench: Beyond Semantic Evaluation for Speech Translation (arXiv:2605.30792v1)
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. OpenSTBench arXiv:2605.30792v1, exact system label GPT Realtime Translate, evaluated as an online streaming speech-to-speech API system and linked by the paper to OpenAI's gpt-realtime-translate model documentation. The benchmark evaluates a system/API route, not a disclosed fixed weight checkpoint. Translation, speech-quality, temporal-consistency, and MSLT metrics use N=1,000 per direction; speaker-preservation metrics use a 300-example paired LibriTTS set; emotion metrics use RAVDESS EN→ZH N=1,440 and MCAE-SPPS ZH→EN N=1,029; nonverbal-event metrics use NonverbalTTS EN→ZH N=359 and SynParaSpeech ZH→EN N=500. The paper does not publish an immutable API serving revision or confidence intervals.
- Context
- Independent evaluator preprint reports a named API/system route. This is not an isolated-model or fixed-checkpoint result: OpenAI's serving revision, backend hardware, decoding controls, and API parameters are not pinned. No numeric confidence interval is reported. The source paper version is arXiv:2605.30792v1 (2026-05-29).
OpenSTBench BLEURT, ZH→EN (Show measurement, test conditions, and source)
- Source value
- 0.4904
- Score
- 0.4904
- Metric
- BLEURT score; higher is better
- Unit
- ratio
- Benchmark version
- OpenSTBench arXiv:2605.30792v1 (Table 5), 2026-05-29
- Direction
- Higher is better
- Sample
- 1000
- Source type
- independent-evaluator
- Evaluator
- OpenSTBench authors, OpenSTBench: Beyond Semantic Evaluation for Speech Translation (arXiv:2605.30792v1)
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. OpenSTBench arXiv:2605.30792v1, exact system label GPT Realtime Translate, evaluated as an online streaming speech-to-speech API system and linked by the paper to OpenAI's gpt-realtime-translate model documentation. The benchmark evaluates a system/API route, not a disclosed fixed weight checkpoint. Translation, speech-quality, temporal-consistency, and MSLT metrics use N=1,000 per direction; speaker-preservation metrics use a 300-example paired LibriTTS set; emotion metrics use RAVDESS EN→ZH N=1,440 and MCAE-SPPS ZH→EN N=1,029; nonverbal-event metrics use NonverbalTTS EN→ZH N=359 and SynParaSpeech ZH→EN N=500. The paper does not publish an immutable API serving revision or confidence intervals.
- Context
- Independent evaluator preprint reports a named API/system route. This is not an isolated-model or fixed-checkpoint result: OpenAI's serving revision, backend hardware, decoding controls, and API parameters are not pinned. No numeric confidence interval is reported. The source paper version is arXiv:2605.30792v1 (2026-05-29).
OpenSTBench UTMOS, EN→ZH (Show measurement, test conditions, and source)
- Source value
- 3.1749
- Score
- 3.1749
- Metric
- UTMOS score; higher is better
- Unit
- rating
- Benchmark version
- OpenSTBench arXiv:2605.30792v1 (Table 6), 2026-05-29
- Direction
- Higher is better
- Sample
- 1000
- Source type
- independent-evaluator
- Evaluator
- OpenSTBench authors, OpenSTBench: Beyond Semantic Evaluation for Speech Translation (arXiv:2605.30792v1)
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. OpenSTBench arXiv:2605.30792v1, exact system label GPT Realtime Translate, evaluated as an online streaming speech-to-speech API system and linked by the paper to OpenAI's gpt-realtime-translate model documentation. The benchmark evaluates a system/API route, not a disclosed fixed weight checkpoint. Translation, speech-quality, temporal-consistency, and MSLT metrics use N=1,000 per direction; speaker-preservation metrics use a 300-example paired LibriTTS set; emotion metrics use RAVDESS EN→ZH N=1,440 and MCAE-SPPS ZH→EN N=1,029; nonverbal-event metrics use NonverbalTTS EN→ZH N=359 and SynParaSpeech ZH→EN N=500. The paper does not publish an immutable API serving revision or confidence intervals.
- Context
- Independent evaluator preprint reports a named API/system route. This is not an isolated-model or fixed-checkpoint result: OpenAI's serving revision, backend hardware, decoding controls, and API parameters are not pinned. No numeric confidence interval is reported. The source paper version is arXiv:2605.30792v1 (2026-05-29).
Page 1 of 4
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-Realtime TranslateFlux 37GPT-4o Mini TTS 36Gemini 3.5 Transcribe Live 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.
Similar models
More models of the same kind as GPT-Realtime Translate, chosen by model line, developer, size, price and release date.
- GPT-Realtime 2.1OpenAI · same developer
- GPT-Live-TranscribeOpenAI · 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
- Nova-3Deepgram · 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
- Subject to limitations. OpenAI lists an inaudible SynthID watermark for supported audio outputs through ChatGPT and the OpenAI API. Coverage can depend on the model, output path and creation date.
- Source: AI watermarking
Licence and access
API
Further details
Prices
- Translation price
- $0.034 per audio minuteSource
Sources and data date
Show all 15 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.