Skip to content
Model infoModelsSnapshot: 15 Oct 2026

Alibaba Cloud / Fun-ASR Realtime

Fun-ASR MTL Realtime · December 2025

Dated multilingual realtime transcription model documented by the SDK and Beijing pricing table.

availableClosed weightsSpeech-to-text
Official documentation (opens in a new tab)Reviewed 14 Oct 2026
14 of 27 specification fields collected
Context windowNot collectedAwaiting source review
Maximum outputNot collectedAwaiting source review
Input billingSee ratesAlibaba Cloud Model Studio · China (Beijing)
Output billingSee ratesPublished units and conditions below

Specifications

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Context / input limit
Not collected
Maximum input tokens
Not collected
Maximum output
Not collected
Reasoning
Not collected
Tool calling
Not collected
Structured output
Not collected
Streaming
SupportedSource for Streaming (opens in a new tab)

WebSocket delivers incremental transcription; audio is sent after task-started.

Batch processing
Not collected
Fine-tuning
Not collected
Knowledge cutoff
Not collected
Release date
Not collected
Parameters
Not collected

The research repository reports 7.7B for its closed Fun-ASR evaluation model and 800M for public Nano models. Exact mapping to this hosted selector is not established.

Architecture
Not collected
Languages
Chinese, English, Japanese, Korean, Vietnamese, Indonesian, ThaiSource for Languages (opens in a new tab)

Chinese dialect/accent coverage is model-specific. Stable Realtime and its February 2026 snapshot have different published language rosters.

Context caching
Not collected
Data retention
Standard API inference prompts and responses are retained up to 30 days, subject to legal/security exceptions; stateful files remain until deleted. Fun-ASR is absent from the published ZDR-supported model roster. Result URL expiry is not a general deletion guarantee.Source for Data retention (opens in a new tab)

Current service policy, not proof of deletion timing for past calls.

Usage metering
Input audio seconds; output text has no separate chargeSource for Usage metering (opens in a new tab)
Request mode
Realtime streaming over WebSocket; AOQ is also documentedSource for Request mode (opens in a new tab)
Audio channels
MonoSource for Audio channels (opens in a new tab)

The speech model audio-specifications table restricts all realtime inputs to mono.

Audio formats
PCM, WAV (PCM), MP3, Opus (Ogg), Speex (Ogg), AAC, AMR-NBSource for Audio formats (opens in a new tab)
Audio sample rates
Specify the actual sample rate; no fixed rate requiredSource for Audio sample rates (opens in a new tab)

Only the 8 kHz selectors require 8,000 Hz. Sample rate is not the model token rate.

Provisional live transcripts
Provisional transcripts may change until sentence completionSource for Provisional live transcripts (opens in a new tab)
Automatic turn detection
VAD or semantic sentence segmentation; configurable silence thresholdSource for Automatic turn detection (opens in a new tab)
Connection behavior
Optional heartbeat keeps the connection alive during continuous silent audioSource for Connection behavior (opens in a new tab)

No exact timeout duration is specified in the reviewed event contract.

Custom vocabulary
Precompiled hotword lists: up to 500 termsSource for Custom vocabulary (opens in a new tab)

Account limit: 10 lists. Inline weighted hotwords and weight-50 super-hotwords are restricted to Qwen-Audio models.

Access & pricing

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

Alibaba Cloud Model Studio · China (Beijing)

Provider source (opens in a new tab)
API model ID
fun-asr-mtl-realtime-2025-12-10
Access
wss://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api-ws/v1/inference; run-task model=fun-asr-mtl-realtime-2025-12-10. Same-region API key; AOQ is an additional documented transport.
Regions
China (Beijing). Deployment scope is not a country-only residency guarantee.
Endpoint availability
Listed in current Model Studio pricingSource for Endpoint availability (opens in a new tab)
Streaming
SupportedSource for Streaming (opens in a new tab)

WebSocket delivers incremental transcription; audio is sent after task-started.

Tool calling
Not collected
Structured output
Not collected
Batch processing
Not collected
Fine-tuning
Not collected
Context caching
Not collected
Included free allowance
No free quotaSource for Included free allowance (opens in a new tab)

Original international-account USD tariff; domestic-account promotions are separate.

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 14 Oct 2026
CategoryUsageRateUnit
Speech recognitionInput audio1$0.000047second of input audio

Rate conditions

  1. Original regional list rate; transcription output has no separate charge. File transcription bills each requested track separately. Free quotas and promotional rates are separate; no generic text-question estimate.
Pricing source (opens in a new tab)

Data & enterprise terms

Standard API inference prompts and responses are retained up to 30 days, subject to legal/security exceptions; stateful files remain until deleted. Fun-ASR is absent from the published ZDR-supported model roster. Result URL expiry is not a general deletion guarantee.

Read the service policy (opens in a new tab)

Caching

Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.

01 / INITIAL INPUTProcess & retain

The system processes eligible input and retains reusable state.

02 / LATER REQUESTCheck available state

Reuse depends on matching input, session state and the cache policy.

03 / CACHE HITReuse & generate

Reusing state can reduce repeated work. Any charges follow the billing rules below.

Alibaba Cloud Model Studio · China (Beijing)

fun-asr-mtl-realtime-2025-12-10
Not verified
How it works
Not collected
Minimum reusable input
Not collected
Lifetime & refresh
Not collected
How to enable it
Not collected
Usage fields
Not collected
Billing rules
Not collected
Limits & data boundaries
Not collected

No verified cache rate for this endpoint. An empty rate is not a zero charge.

Caching reviewed 14 Oct 2026

What the measurements say

No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.

Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.

The record behind the model

Why it is included

The official Fun-ASR repository explicitly labels its evaluated 7.7B Fun-ASR model as not open-source, separately from public 800M Nano checkpoints. Hosted release cards and API selectors establish these service identities. Extending family-level closed-weight classification to each private revision is an inference; exact research-checkpoint mapping and parameter counts are not copied onto every hosted release.

Classification evidence (opens in a new tab)

Sources reviewed