🔍 Intelligence Synthesis JP/EN
Ethics · Regulation · Technology — Pharma Practice Notes

AI Tech Flash ── 2026-11-18

DeepSeek API adds search, OCR and speech

DeepSeek API adds search, OCR and speech; audio is capped at 120 seconds

What changes

According to Tech Times, developers can now send audio, documents, web links and search queries through one vendor. DeepSeek's changelog entry is dated Nov. 13.

Teams building agents or retrieval-augmented generation (RAG, where a model looks up outside documents to answer) can get search, document parsing and voice from one provider.

Why it matters

Tech Times says this consolidation puts audio, files, fetched pages and search queries on one data path. The changelog does not discuss that concentration, the outlet says.

DeepSeek is a Hangzhou-based company subject to China's state-security laws. Tech Times says developers should weigh this before they compare per-unit prices.

What it is

DeepSeek's changelog lists web search, URL reading, file input and a safety check setting. It also lists standalone search, fetch, OCR, speech recognition (ASR) and speech synthesis (TTS) services.

OCR reads text from images and documents. ASR turns speech into text, and TTS turns text into speech. The update also includes an open-source demo chat client that can import a user's DeepSeek chat history.

Who and where

The vendor is DeepSeek, based in Hangzhou, China. Tech Times reports that the chat features run on deepseek-flash across the Messages, Responses and Chat Completions APIs.

The pages for OCR, speech recognition and speech synthesis name no underlying model. The documentation describes the tools only as hosted services, with no self-hosted option.

Why it was built

The changelog does not appear to state a purpose in detail. From Tech Times, the implied aim is to cut the effort of gathering search, document parsing and voice from separate vendors.

The article frames this as developer convenience. It does not quote DeepSeek as stating that reason.

How it works

Tech Times, citing TokenPost, says DeepSeek's servers handle tool calling and context assembly. Developers display the content blocks the API returns and pass them back in the next request.

Files are uploaded first, parsed, then referenced by a file ID in a chat message. Text recognition mode extracts text only. Rich layout mode crops figures into the text, but DeepSeek warns that not every figure is guaranteed to be found.

Speech recognition accepts 16-bit PCM WAV or Ogg Opus files of up to 120 seconds and 8 MiB. It does not support streaming, and for stereo files only the left channel is recognized.

Key points

  1. Speech recognition takes up to 120 seconds and 8 MiB per request, and reads only the left stereo channel.
  2. DeepSeek publishes no accuracy figures for OCR, speech or search, and only one outside test of search was found.
  3. DeepSeek is subject to China's state-security laws, so user data sits under that legal jurisdiction.

Conclusion

DeepSeek's API now offers search, OCR and speech from one vendor. Tech Times says the limits, the thin outside testing and the Chinese legal exposure should be checked first. It advises keeping sensitive audio and contracts out of the hosted tools.

Quotes and sources

“its speech recognition service accepts only 120 seconds of audio per request, its file parser can miss charts and diagrams, and every call goes to a Hangzhou-based company subject to China's state-security laws.”
── Tech Times, 2026-11-16
“Speech recognition is the tightest constraint.”
── Tech Times, 2026-11-16
“DeepSeek's pages publish no accuracy figures for the new tools.”
── Tech Times, 2026-11-16
“Treat DeepSeek's capability claims for the new tools as unverified until a named third party tests them on your own audio and documents.”
── Tech Times, 2026-11-16

Source articles

Written by AI (Claude) from the cited articles, then checked by Jev (TypeSafe's judgement model), a Gemini fact-check and a verbatim check of every quote.