Compare speech and transcription model evidence.
Speech recognition, speech generation and other audio tasks need different checks. This page separates models explicitly marked for transcription from the rest of the audio catalog, with the recorded price unit and source beside each model.
65 models explicitly list transcription; 293 other models are classified as audio. Each group shows up to six. Audio classification alone does not establish speech recognition, synthesis, streaming or language coverage.
Catalog snapshot · 12 rows shownConfirm the direction of the task
A transcription flag is explicit catalog evidence for speech-to-text. The broader audio label can also cover speech synthesis, music or other audio work. It does not establish that a model accepts speech and produces a transcript. Open the model profile and source to check its exact task.
For transcription, test your language, accents, background noise and recording format. Check timestamps, speaker labels and streaming in the provider documentation if your application needs them. These features are not inferred from an audio-input flag or a model name.
Keep the clock and text units separate
A per-hour or per-minute quote usually needs a duration quantity; a character quote needs the billable text definition. Token and generation rates use their own quantities. The table preserves each unit and configuration rather than converting all audio models to a single cost per minute.
A recorded rate can be incomplete, conditional or overdue. Use its status and check date before estimating a bill. Benchmark word error rate, when you encounter it on a model page, is a lower-is-better error measure. It is not interchangeable with a general audio quality score.
How this view is assembled
Selection: up to six models with an explicit transcription capability and up to six other audio models. Models with a current eligible route appear first, then names. One recorded route per model is shown, favoring current quotes and stable provider order. Missing task, price or feature evidence remains unknown.
Models with transcription evidence
Listed task evidence and one recorded quote per model; billing units remain separate
| Model | Task evidence | Recorded API quote | Sources and dates |
|---|---|---|---|
| AssemblyAI: Universal-3.5 Pro | Transcription explicitly listedAudio input listed | OpenRouter$83.3333 input / $0.00 outputPer 1M tokens · USDConfiguration not reportedCurrent eligible quoteCurrent eligible quote | OpenRouter ↗Catalog updated Open ASR Leaderboard ↗Catalog updated OpenRouter price source ↗Checked |
| Mistral: Voxtral Small 24B 2507 | Transcription explicitly listedAudio input listed | DeepInfra$0.00005 / secondUSDInput audio secondCurrent eligible quoteCurrent eligible quote | OpenRouter ↗Catalog updated OpenRouter ↗Catalog updated DeepInfra price source ↗Checked |
| NVIDIA: Parakeet TDT 0.6B v3 | Transcription explicitly listedAudio input listed | OpenRouter$25.00 input / $0.00 outputPer 1M tokens · USDConfiguration not reportedCurrent eligible quoteCurrent eligible quote | OpenRouter ↗Catalog updated Open ASR Leaderboard ↗Catalog updated OpenRouter price source ↗Checked |
| OpenAI: GPT Transcribe | Transcription explicitly listedAudio input listed | OpenRouter$75.00 input / $0.00 outputPer 1M tokens · USDConfiguration not reportedCurrent eligible quoteCurrent eligible quote | OpenRouter ↗Catalog updated Open ASR Leaderboard ↗Catalog updated OpenRouter price source ↗Checked |
| Voxtral-Mini-3B-2507 | Transcription explicitly listedAudio input listed | DeepInfra$0.0000166667 / secondUSDInput audio secondCurrent eligible quoteCurrent eligible quote | OpenRouter ↗Catalog updated DeepInfra ↗Catalog updated DeepInfra price source ↗Checked |
| whisper-large-v3 | Transcription explicitly listedAudio input listed | DeepInfra$0.0000075 / secondUSDInput audio secondCurrent eligible quoteCurrent eligible quote | OpenRouter ↗Catalog updated DeepInfra ↗Catalog updated DeepInfra price source ↗Checked |
Other audio models to inspect by task
Listed task evidence and one recorded quote per model; billing units remain separate
| Model | Task evidence | Recorded API quote | Sources and dates |
|---|---|---|---|
| acestep-v15-xl-sft | Audio modality; exact task must be checkedAudio input not separately reported | DeepInfra$0.001 / secondUSDOutput second$0.001 per second of generated musicCurrent eligible quoteCurrent eligible quote | DeepInfra ↗Catalog updated DeepInfra price source ↗Checked |
| Audio8-TTS-Preview-0.6b | Audio modality; exact task must be checkedAudio input not separately reported | DeepInfra$0.005 / 1K charactersUSD1,000 input charactersCurrent eligible quoteCurrent eligible quote | DeepInfra ↗Catalog updated DeepInfra price source ↗Checked |
| Bytedance Seed Speech Text to Speech | Audio modality; exact task must be checkedAudio input not separately reported | fal$0.03 / 1K charactersUSDConfiguration not reportedCurrent eligible quoteCurrent eligible quote | fal ↗Catalog updated fal price source ↗Checked |
| Chatterbox | Audio modality; exact task must be checkedAudio input not separately reported | fal$0.015 / minuteUSDConfiguration not reportedCurrent eligible quoteCurrent eligible quote | fal ↗Catalog updated fal price source ↗Checked |
| Chatterbox | Audio modality; exact task must be checkedAudio input not separately reported | fal$0.025 / 1K charactersUSDConfiguration not reportedCurrent eligible quoteCurrent eligible quote | fal ↗Catalog updated fal price source ↗Checked |
| Chatterbox | Audio modality; exact task must be checkedAudio input not separately reported | fal$0.025 / 1K charactersUSDConfiguration not reportedCurrent eligible quoteCurrent eligible quote | fal ↗Catalog updated fal price source ↗Checked |
Where these rows come from
We join catalog specifications, recorded provider quotes and published evaluation records where exact model identities are available. Some observations come through aggregators or benchmark collections. Multiple listings do not establish independent confirmation of a claim.
Open the source beside a row to check the original evidence. A collection or check date is separate from a model release or evaluation date. Unknown values stay unknown. Recorded prices are in USD with their original billing units and settings.
See data coverage and sources →Compare for your own use case
Keep your task, model identity and provider configuration together. Explore the catalog yourself, or connect your AI to read the same source-backed records.