# Baseten — The fastest, lowest-cost batch transcription with Qwen3-ASR

- Company: Baseten (baseten.co)
- Announced: 2026-10-28T17:39:04+00:00
- Category: infrastructure-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.baseten.co/blog/fastest-lowest-cost-qwen3-asr-batch-transcription
- Record: https://forck.live/items/28878-the-fastest-lowest-cost-batch-transcription-with-qwen3-asr
- Subject: Baseten
- Open weights: yes
- Models affected: Qwen3-ASR 1.7B
- License: Apache 2.0
- Pricing: about 1/10th of a cent per audio hour

Baseten has optimized Qwen3-ASR 1.7B for batch transcription, claiming the lowest cost per audio minute across all transcription providers. The Qwen team's open-weight speech recognition model is the top-ranking open-weight model on Hugging Face's Open ASR Leaderboard, with a 5.4% average word error rate, and Baseten says its real-time deployment also has the lowest WER and latency on Coval's speech-to-text benchmark. Batch transcription is a throughput problem rather than a latency one, so Baseten profiled every stage of a request under load and changed four things: native audio decoding, KV cache quantization, host-to-device syncs and request handling. Replacing vLLM's PyAV loader with torchcodec hands whole files to FFmpeg in one native call, and long files are decoded in parallel across time ranges, so an hour of audio is preprocessed in well under a second. Storing the KV cache in FP8 halves the bytes read each step and doubles the sequences that fit in memory, with FlashInfer kernels used because vLLM's default FlashAttention-2 backend does not support FP8 cache on RTX PRO 6000; weights stay in BF16 because quantizing them raised word error rate. Keeping host-side copies of sequence lengths and clip lengths removed synchronization points, lifting GPU utilization under load from 86% to 97%. On hour-long files Baseten says it sustains over 3,000x real-time factor at about 1/10th of a cent per audio hour, with 7x more audio per second than out-of-the-box vLLM and no meaningful change in word error rate.

## Evidence

Verbatim from https://www.baseten.co/blog/fastest-lowest-cost-qwen3-asr-batch-transcription:

> Transcribe an hour of audio with Qwen3-ASR for 1/10 of a cent, with 7x higher throughput, 6.9x lower latency, and the lowest word error rate.
> 
> Authors
> 
> Published
> 
> October 28, 2026
> 
> Share
> 
> Published
> 
> October 28, 2026
> 
> Qwen3-ASR 1.7B is the top-ranking open-weight model on Hugging Face's Open ASR Leaderboard , but serving it efficiently for batch transcription requires careful optimization. We combined native audio decoding, KV cache quantization, and reduced unnecessary host-to-device syncs to serve 7x more audio per second at high concurrency on a single RTX PRO 6000. On hour-long files, we sustain over 3,000x real-time factor (RTF) at about 1/10th of a cent per audio hour, with no meaningful change in word error rate.
> 
> Speech-to-text splits into two very different workloads. On the one hand, real-time transcription is about latency: a voice agent needs words back within a few hundred milliseconds of someone speaking. On the other, batch transcription is about throughput: podcast networks, meeting-notes products, and call-center archives have hours of recorded audio and want it transcribed as cheaply as possible.

---

Record: https://forck.live/items/28878-the-fastest-lowest-cost-batch-transcription-with-qwen3-asr
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md