Qwen: Qwen3 ASR Flash

qwen/qwen3-asr-flash-2026-02-10

Qwen3-ASR-Flash is Alibaba's automatic speech recognition service, built on the Qwen3-Omni foundation and trained on tens of millions of hours of multimodal speech data. The model handles 11 languages — including Chinese (with Cantonese, Sichuanese, Minnan, and Wu dialects), English, Arabic, French, German, Spanish, Italian, Portuguese, Russian, Japanese, and Korean — with automatic language detection so no manual configuration is needed for mixed-language audio.

The model is designed for difficult acoustic conditions: it transcribes lyrics over background music, handles noisy and far-field recordings, filters silence and non-speech audio, and accepts arbitrary context text (names, jargon, domain terminology) to bias recognition toward specific vocabulary.

Modalities

Price

$0.000035/second

Released

May 14, 2026

Providers

This model is hosted by one provider. OpenRouter forwards every request to it directly — no routing decisions to make.

Effective Pricing

The chart below shows the average price customers are actually paying after prompt caching. Depending on the amount of repeated context you send, this can be 60–80% cheaper than the provider list price. Shown are rolling averages from the past 30 days.

Performance

Throughput is how fast the model writes (tokens per second — higher is better). Latency is total round-trip time (lower is better). TTFT is time-to-first-token — how long before you see anything appear (lower is better).

Uptime

Percent of requests that succeeded over the last 30 days. OpenRouter monitors every provider continuously and automatically retries on the next-best provider when one returns an error.

Apps

Public apps that send the most traffic to this model. Good signal for what real production workloads look like — and a hint at which use cases this model is best suited for.

Activity

Token volume and request traffic to this model over time.

Quick Start

Drop-in code to call this model. OpenRouter's API is OpenAI-compatible — most SDKs work by just swapping the base URL. The only thing that changes between models is the model slug below.