Speech-to-text model pricing compared
Per-hour cost of every transcription model BibiGPT can run, plus a few reference APIs. Token-billed models are converted with the densities we measured.
Prices checked 2026-10-26. These are upstream list prices, not BibiGPT pricing.
| Model | per hour | 10 h | List price | Strengths | In BibiGPT |
|---|---|---|---|---|---|
Bibi Local on-device transcription Bibi Local desktop app SenseVoice / Whisper (on device)Audio never leaves your machine. Speed depends on your computer. | $0 | $0 | Free (runs on your computer) | zh/en/ja/ko/yue and more Word timestamps | On device Desktop app, local mode Download Bibi Local |
Qwen-Audio-3.1 ASRCheapest Alibaba Qwen qwen-audio-3.1-asr-flash-filetransAbout ¥0.075–0.10 per hour on our Chinese samples, the cheapest cloud option we measured. Better on product names than Paraformer; filler words are kept. | $0.0136Measured | $0.136 | ¥0.8 input / ¥2.7 output per 1M tokens (Beijing region) | 30 languages + 10 Chinese dialects Word timestampsSpeaker labelsHotwordsDialects | In BibiGPT |
StepAudio 2.5 ASR StepFun stepaudio-2.5-asrThe cheapest duration-billed API. Strong on Mandarin; no dialects and no speaker labels. | $0.0211 | $0.211 | ¥0.15/hour ($0.022/hour) | Chinese, English Word timestampsHotwords | In BibiGPT |
Paraformer v2 Alibaba Qwen paraformer-v2BibiGPT’s default for Chinese video, with speaker labels and filler-word removal on. Weak on brand names. | $0.0406 | $0.406 | ¥0.00008/second | zh/en/ja/ko/de/fr/ru + ~18 accents Word timestampsSpeaker labelsHotwordsFiller removalDialects | In BibiGPT |
DeepSeek POST /v1/tools/asrMeasured 1.9% CER on FLEURS Chinese (2nd of 6) and 6.4% WER on English, about 1 s per clip. Max 120 s per file and no timestamps, so it suits voice input rather than subtitles. | $0.058 | $0.58 | $0.116/hour (limited-time 50% off: $0.058), billed per second | 21 languages incl. Cantonese, auto-detected Dialects | Not in BibiGPT |
Grok Voice Transcribe 2.0 xAI grok-voice-transcribe-2.0A cheap pick for English and other non-CJK audio. Not for Chinese. | $0.1 | $1 | $0.10/hour (batch) | 25 languages (no CJK) Word timestampsSpeaker labels | In BibiGPT |
Whisper large-v3 (Groq) OpenAI open source / Groq whisper-large-v3Most languages and very fast. No speaker labels; can hallucinate on silence. | $0.111 | $1.11 | $0.111/hour | 99 languages Word timestamps | In BibiGPT |
Qwen3-ASR Flash Alibaba Qwen qwen3-asr-flash-filetransReturns emotion labels, which 3.1 does not. No speaker labels. | $0.112 | $1.12 | ¥0.00022/second | ~27 languages + 4 dialects Word timestampsEmotionDialects | In BibiGPT |
Qwen-Audio-3.0 ASR Alibaba Qwen qwen-audio-3.0-asr-flash-filetransWider dialect list than 3.1, at roughly 8–10× the price. | $0.112 | $1.12 | ¥0.00022/second | Widest Chinese dialect coverage Word timestampsDialects | In BibiGPT |
ElevenLabs Scribe v2 ElevenLabs scribe_v2Reliable across languages with speaker labels. Bring your own ElevenLabs key. | $0.22 | $2.2 | $0.22/hour (batch) | 90+ languages Word timestampsSpeaker labelsSound eventsHotwords | In BibiGPT · your key |
Deepgram Nova-3 Deepgram deepgram/nova-3A long-standing pick for English calls and meetings. Selectable with an OpenRouter key. | $0.258 | $2.58 | $0.0043/minute (OpenRouter) | Multilingual, strongest in English Word timestampsSpeaker labelsHotwords | In BibiGPT · your key |
GPT Transcribe OpenAI openai/gpt-transcribeSelectable with an OpenRouter key. Text only, no timestamps or speakers. | $0.27 | $2.7 | $0.0045/minute (OpenRouter) | Multilingual | In BibiGPT · your key |
Gemini 3.5 Transcribe Google gemini-3.5-transcribeRemoves fillers and self-corrections natively. Bring your own Gemini key. | $0.31Est. | $3.1 | $2 audio input / $12 text output per 1M tokens (official estimate about $0.005/minute) | 85+ languages Word timestampsSpeaker labelsFiller removal | In BibiGPT · your key |
Doubao recorded-audio ASR ByteDance Volcengine volc.bigasr.aucStrong on Chinese and dialects. Needs a Volcengine AppID and token. | $0.324 | $3.24 | Standard ¥2.3/hour; off-peak ¥1.2; file ASR 2.0 ¥0.8 | zh/en + 5 dialects Word timestampsSpeaker labelsHotwordsDialects | In BibiGPT · your key |
MAI-Transcribe 1.5 Microsoft microsoft/mai-transcribe-1.5Selectable in the transcriber with an OpenRouter key. | $0.36 | $3.6 | $0.36/hour (OpenRouter) | Multilingual Word timestamps | In BibiGPT · your key |
Fish Audio Transcribe Pro Fish Audio transcribe-1-proSpeaker and sound-event tags; BibiGPT strips events like [click] from subtitles. Average on Chinese names. | $0.36 | $3.6 | $0.0001/second | 83 languages (vendor claim) Word timestampsSpeaker labelsSound events | In BibiGPT · your key |
StepAudio 3 ASR StepFun stepaudio-3-asr-maxTied first on Artificial Analysis batch WER. Without hotwords it heard “BibiGPT” as “Bilibili GPT”. | $0.394 | $3.94 | ¥2.8/hour (international site $0.40/hour) | Chinese, English Word timestampsHotwords | In BibiGPT · your key |
- Gemini audio input: 25 tokens per second; transcript output about 3 tokens per second.
- Qwen-Audio-3.1: about 20 input + 4 output tokens per second, measured on our own Chinese samples.
- CNY ↔ USD at 7.1.
FAQ
Which transcription model is cheapest?
On our samples Qwen-Audio-3.1 file transcription came to about ¥0.075–0.10 per hour, below StepAudio 2.5 (¥0.15/h) and Paraformer v2 (¥0.288/h). On-device models in the Bibi Local desktop app cost nothing.
What is the difference between STT, live and TTS?
STT turns a finished recording into text and is billed per audio hour. Live models do the same while you speak, or answer you in voice, and cost more. TTS goes the other way: text in, speech out, usually billed per character.
Why are some prices marked “Est.”?
Those models bill by tokens, not minutes. We convert with the token densities listed under the table; your real bill moves with speech speed and language.
Can I use these models in BibiGPT?
Models marked “In BibiGPT” are covered by membership. “Your key” models are selectable once you paste your own API key. “On device” models run free in the Bibi Local desktop app. “Not in BibiGPT” rows are listed for reference.
Try them on your own audio
Paste a YouTube, Bilibili or podcast link into BibiGPT and pick the engine in the transcription settings. For private recordings, Bibi Local transcribes on your own computer.