Intel / post

StepAudio 2.5 ASR: 30-minute inputs at $0.022 per hour

Source: x.com
Date: Published Apr 24, 2026

StepFun@StepFun_ai

StepAudio 2.5 ASR is here! 4x faster decoding. 30-minute continuous audio. $0.022/hour. We built this to push ASR further — faster inference, longer inputs, bilingual accuracy that holds up, and a price point that actually scales. ⚡ 4x decoding speed vs. previous generation ⚡ 30-min continuous audio processing ⚡ Top-tier bilingual (EN/ZH) performance ⚡ 80% lower inference cost — API at $0.022/hr For more details, check out: - Documentation: https://t.co/oiya8DpmfM - Demos: https://t.co/9ob48LOGqs - Model card: https://t.co/xo1YDteuQG

Apr 24, 2026 View on X

Context

StepFun released StepAudio 2.5 ASR, a speech-to-text model the company says decodes 4 times faster than its previous generation, handles 30 minutes of continuous audio without breaking it into chunks, and prices API access at $0.022 per hour, which StepFun describes as an 80% cost cut from the earlier model.

StepFun also claims strong bilingual accuracy across English and Chinese, but did not publish a word-error-rate figure or independent comparison alongside this announcement, so the accuracy claim is StepFun's characterization rather than a measured number. The pricing and processing-length figures are concrete and checkable against the linked documentation and model card.

Terms in this piece · Glossary

benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.

Read the source x.com