DeepSeek Opens Full Suite of Application APIs, Speech Recognition Costs Under $0.06 Per Hour
nashnova research
DeepSeek opened its entire application API suite — web search, speech recognition, speech synthesis and more — for independent, pay-per-use access. Speech recognition lands at ¥0.39/hour after discount, less than one-fifth of OpenAI's comparable service, sharply lowering the cost floor for developers building AI apps.
What just opened, and how is it different from before?
Features previously locked inside DeepSeek's own app — web search, deep-thinking search, file upload, link reading — are now available as standalone APIs, individually callable and individually billed.
The biggest addition is voice: DeepSeek is offering official speech-recognition (ASR — turning audio into text) and speech-synthesis (TTS — turning text into audio) endpoints for the first time.
This means → developers no longer need to stitch in third-party voice services. One DeepSeek API set can power a full voice-interactive application.
How much cheaper is it, exactly?
Speech recognition lists at ¥0.78/hour; a 50 % launch discount brings it to ¥0.39/hour, roughly ¥0.0065 per minute. Speech synthesis is ¥1.5 per 10 000 characters after discount.
Three mainstream comparisons: OpenAI's GPT-Transcribe costs about ¥1.81/hour; the budget GPT-4o mini Transcribe runs roughly ¥1.21/hour; Google Cloud's standard speech recognition comes to about ¥6.43/hour.
In plain terms = DeepSeek's discounted price is 1/4.6, 1/3.1, and 1/16.5 of those three, respectively — at the extreme end, Google charges 16 times as much for the same job.
What changed with web search?
The web-search API is not new — it launched on October 17 this year. But back then, developers had to write their own orchestration code to chain search decisions, page fetching, and model summarization.
After this upgrade, developers just flip a switch. When to search, what to search for, and how to integrate results are all handled by DeepSeek automatically.
This means → the integration bar dropped from "build an orchestration pipeline" to "toggle one setting." Small and mid-size dev teams will see noticeably shorter build cycles.
Why does this look like DeepSeek is handing the application layer away?
DeepSeek stated explicitly that developers can use these APIs to "replicate the full DeepSeek App 1:1," and open-sourced a lightweight chat-frontend demo alongside the announcement.
The company's own words: "DeepSeek's current conversational product is just a relatively basic form."
This reflects DeepSeek's consistent strategy: invest in the model layer, don't deeply operate consumer products. Opening full capabilities at rock-bottom prices hands the question of "what does the next hit app look like?" to the external ecosystem — whether one emerges is the key test of this playbook.
市场有风险，内容仅供研究参考，不构成投资建议。