Introducing Qwen3-ASR-Flash: Top-Tier AI Transcription with Leaderboard-Leading Accuracy

We've added a second transcription pipeline: Qwen3-ASR-Flash from Alibaba — ranked #4 of 80+ on the HuggingFace Open ASR Leaderboard with a 6.37% average Word Error Rate. It runs alongside our existing GPT-4o Transcribe Diarize pipeline.

Both pipelines include the same downstream AI analysis — sections, chapters, summaries, semantic search, and Q&A with citations. The difference is in the transcription step itself.

Why Two Pipelines?

Not every transcription job needs the same model. A 3-hour lecture by a single speaker has different needs than a 20-minute podcast with 4 guests.

• GPT-4o Transcribe Diarize is the best choice when you need speaker identification (diarization).
• Qwen3-ASR-Flash is the best choice for top-tier accuracy, long single-speaker audio, word-level timestamps, or Chinese dialect support.

Head-to-Head Comparison

Where the Cost Difference Comes From

The Qwen3 transcription API is priced lower than GPT-4o ($0.13/hr vs $1.80/hr). The rest of the pipeline (LLM processing, semantic search embeddings, and infrastructure) is shared between both pipelines.

When to Use Each Pipeline

Choose GPT-4o Diarize when:

• You have multiple speakers (meetings, interviews, podcasts). For podcast transcription best practices, see the podcast guide.
• You need speaker labels in your transcript
• Maximum transcription accuracy matters most
• You're working in languages where GPT-4o excels

Choose Qwen3-ASR-Flash when:

• You want leaderboard-leading accuracy (#4 of 80+ on HuggingFace Open ASR Leaderboard)
• You have long recordings (lectures, webinars, audiobooks) — try chapters for long recordings
• The audio has a single speaker or you don't need speaker labels
• You need word-level timestamps for precise subtitle generation
• You need Chinese dialect support (Cantonese, Sichuanese, and 20 more)
• You want emotion detection in the transcript

Chinese Dialect Support

Qwen3-ASR-Flash stands out with support for 22 Chinese dialects including Cantonese, Sichuanese, Fujian, Henan, Hubei, and more — far beyond what most transcription services offer. If your content includes regional Chinese speech, this pipeline is a significant upgrade.

Same AI Analysis, Different Starting Point

Regardless of which pipeline you choose, every transcription gets the same enrichment:

• Section detection using advanced NLP
• Chapter generation with titles and summaries
• Semantic search across your transcript
• Q&A with citations that link to exact timestamps
• AI summary with takeaways, quotes, and speaker profiles

How to Use It

1. Go to Transcribe
2. In the pipeline selector, choose Qwen3-ASR-Flash
3. Upload your file or paste a YouTube URL
4. Review the quote — you'll see the lower rate automatically

You can switch between pipelines for each transcription. Use GPT-4o for your podcast interviews and Qwen3 for your lecture recordings.

Pricing with Plans

Pricing is flat, not per-minute. Qwen3-ASR-Flash runs on our self-hosted engine, so transcription with it is unlimited on every paid plan. GPT-4o Diarize is a premium model, so it is pay-as-you-go from your wallet.

Annual billing gives you 2 months free, and every new account starts with a $5 signup credit. Premium-model usage draws from your wallet at pay-as-you-go rates.

Try Qwen3-ASR-Flash today at transcribe.so/transcribe. Choose the pipeline that fits your job — speaker diarization with GPT-4o, or leaderboard-leading accuracy with Qwen3.