As of: 2026-10-25 23:49 CET.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/
Canonical page: https://postcutoff.com/m/fun-realtime-asr/

# Fun-Realtime-ASR-preview (Fun-ASR-Flash / Fun-ASR-Realtime)

Alibaba Cloud (Tongyi Lab), Fun-ASR family. Status: current. Type: speech and audio. Released: 2026-07-06.

## How to call it

| Provider | Model id | Endpoint | Docs |
|---|---|---|---|
| Alibaba Cloud Model Studio (offline, HTTP) | `fun-asr-flash-2026-06-15` |  | https://www.alibabacloud.com/help/en/model-studio/fun-asr-flash |
| Alibaba Cloud Model Studio (streaming, WebSocket) | `fun-asr-realtime` | wss://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api-ws/v1/inference | https://www.alibabacloud.com/help/en/model-studio/fun-asr-realtime-python-sdk |
| Alibaba Cloud Model Studio (model list) |  |  | https://www.alibabacloud.com/help/en/model-studio/asr-model/ |

## Pricing

- Flash per second intl: $0.0000
- Flash per second beijing: $0.0000
- Realtime per second intl: $0.0001
- Realtime per second beijing: $0.0000
- Unit as published: USD per second of input audio, list prices excl. promotions (fun-asr-flash-2026-06-15 from its model page; fun-asr-realtime from the Model Studio pricing page). Artificial Analysis lists $2.82 per 1,000 min, which equals $0.000047/s (the Beijing fun-asr-realtime price)
- Source: https://www.alibabacloud.com/help/en/model-studio/model-pricing
- Last checked: 2026-10-25

## Specs

- Input: audio
- Output: text
- Context window: not published
- Max output: not published
- Open weights: no
- Licence: proprietary

## Where it stands

- Newer speech and audio models in the registry since its release: 46. The 5 newest:
  - Eleven Realtime v1 Max / Mini (2026-10-20) https://postcutoff.com/m/eleven-realtime-v1-max/
  - Voxtral Mini Transcribe 3 (+ Mini Transcribe Realtime 3) (2026-10-20) https://postcutoff.com/m/voxtral-mini-transcribe-3/
  - Voxtral Small Transcribe 3 (2026-10-20) https://postcutoff.com/m/voxtral-small-transcribe-3/
  - Voxtral TTS 3 (Voxtral Mini TTS 3) (2026-10-20) https://postcutoff.com/m/voxtral-mini-tts-3/
  - GPT-Live-1.1 (2026-10-15) https://postcutoff.com/m/gpt-live-1-1/
- Closed weights; access is hosted only.

## What stands out

- #1 on Artificial Analysis AA-WER (1.7%): Listed by Artificial Analysis as Fun-Realtime-ASR-preview, #1 on its non-streaming speech-to-text leaderboard with AA-WER 1.7% (1.8% on long-form Earnings22), speed factor ~20x real time, $2.82 per 1,000 min (read 2026-10-25). Source: https://artificialanalysis.ai/speech-to-text
- One model for 30 languages and 16 Chinese dialects: July 2026 upgrade: single streaming model for 30 languages and 16 dialects, 88.62% average character accuracy across the dialects (Shanghainese 92.41%), first-word latency in the hundreds of milliseconds, hot words and conversation-history context (Alibaba figures via IT Home). Source: https://www.ithome.com/0/973/132.htm
- Prompt-context offline transcription: fun-asr-flash-2026-06-15 accepts prompt context, covers 30 languages plus the main Chinese dialect groups and 20+ regional Mandarin accents; files up to 5 minutes. Source: https://www.alibabacloud.com/help/en/model-studio/fun-asr-flash

## Notes

Artificial Analysis speech-to-text (https://artificialanalysis.ai/speech-to-text), read 2026-10-25: Fun-Realtime-ASR-preview rank #1, AA-WER 1.7%, $2.82 per 1,000 min. Name mapping is medium confidence: AA never gives an API id; IT Home (2026-07-06) says the AA entry 'Fun-realtime-ASR-preview' is the former name of the offline Fun-ASR-Flash (fun-asr-flash-2026-06-15), and AiCoin reports Fun-ASR-Flash took the top AA-WER spot, but AA's $2.82/1k min matches the Beijing fun-asr-realtime price, not fun-asr-flash. Alibaba's docs do not use the name 'Fun-Realtime-ASR' or label any Fun-ASR model preview, so status is set to current (caller had proposed preview, after AA's name). Other ids: fun-asr-realtime-2026-02-28 (zh/en/ja + dialects), fun-asr-mtl-realtime-2025-12-10, fun-asr-flash-8k-realtime(-2026-01-28) for 8 kHz telephony, offline fun-asr (diarization, 12 h files). fun-asr-realtime international price includes 10 free hours for 90 days. No Beijing endpoint given on the SDK page. Released date = upgrade announcement 2026-07-06 (IT Home); the fun-asr-realtime line dates back to 2025-09-15 snapshots. No timeline entry yet; related: qwen-audio-3-1-asr, fun-cosyvoice3.

Alibaba's Fun-ASR speech recognition (Bailian / Model Studio), top of Artificial Analysis' word-error-rate ranking in October 2026.

Sources: https://www.alibabacloud.com/help/en/model-studio/asr-model/ · https://www.alibabacloud.com/help/en/model-studio/fun-asr-flash · https://www.alibabacloud.com/help/en/model-studio/model-pricing · https://www.alibabacloud.com/help/en/model-studio/fun-asr-realtime-python-sdk · https://www.ithome.com/0/973/132.htm · https://www.aicoin.com/en/news-flash/2969731 · https://artificialanalysis.ai/speech-to-text

Model page: https://postcutoff.com/m/fun-realtime-asr/