DeepSeek has opened the tools behind its consumer chat app to developers, according to an entry dated 13 November in its API change log: “Today we open these capabilities to developers”. Three are newly priced standalone APIs: OCR at $0.225 per 1,000 pages, speech to text at $0.116 per hour of audio and text to speech at $0.45 per 10,000 characters, with both speech prices halved for a limited time.
At those prices, recognising 100,000 pages costs $22.50 and transcribing 1,000 hours of audio costs $58 at the limited-time price, by our arithmetic. The speech-to-text API takes clips of at most 120 seconds, so an hour of audio means at least 30 requests. Search and Fetch, which DeepSeek opened in October, stay at $0.75 per 1,000 requests each.
Business Pill · The Model and the Harness
An evergreen Business Engineer explainer on why products built on the same model can differ: the harness around the model. It teaches the concept only and does not discuss DeepSeek.
The key insight: As we read the change log and the price list together, DeepSeek is now selling, as separate metered services, the pieces a developer would otherwise assemble around a model to build a chat product: search, page reading, document recognition, voice in and out, and a safety review. Search and Fetch kept their October price; what is new on 13 November is that the rest of the app has a price per page, per hour or per character.
The change log says that in the DeepSeek App, “web search, file input, URL reading, voice interaction and safety work together with model inference to form the conversation experience users know”. The “deepseek-flash” model now supports web search, deep thinking search, file input, URL reading and a model safety check in the Messages, Responses and Chat Completions APIs, with “the tool calls and context assembly handled by the service”.
The underlying capabilities and the voice features are also offered as standalone APIs: Search and Fetch, OCR, ASR (speech to text) and TTS (text to speech). DeepSeek says it has “built and open-sourced a demo chat client on these APIs, which can import your DeepSeek chat history”; the chat-demo repository on GitHub was created on 12 November and carries the MIT licence.
Cost of four sample workloads at the prices on DeepSeek’s pricing page on 13 November 2026, our arithmetic; the workload sizes are ours, for illustration. Speech prices are DeepSeek’s limited-time prices (50% off), with list prices in brackets. Model tokens are billed separately and are not included.Entries in DeepSeek’s API change log in 2026, with prices from its pricing page on 13 November. Speech prices marked with an asterisk are DeepSeek’s limited-time prices, 50% off list. The chat demo’s licence is from its GitHub repository.
The Price List and the Limits
OCR and file parsing. Both are billed at $0.225 per 1,000 pages, according to the pricing page. The OCR API accepts PDF, Word, PowerPoint and image files up to 100 MiB and 1,000 pages per request, so the largest single request costs $0.225, our arithmetic. The usage tokens it returns are “for reference only; not used for billing”.
Speech. Speech to text lists at $0.116 per hour of audio and text to speech at $0.45 per 10,000 characters, each shown with a “Limited-time price (50% off)” of $0.058 and $0.225. Speech to text is billed to the second, covers 21 languages detected from the audio, and does not support streaming. Text to speech offers two voices, Stella and Tide, in 10 languages, with up to 10,000 billed characters per request and a WebSocket mode for streaming.
Search and safety. The Search and Fetch guide says calls made through DeepSeek Harness, Claude Code or Codex “are also billed per successful request”; our October piece on the search API reported a free period for those calls under the documentation at the time. The safety check bills its review tokens as output tokens, and DeepSeek tells developers serving the public to enable it and “fulfill obligations such as content safety review, large model registration and AI labeling”.
The Structural Read
As we read the limits, the products are built for chat turns more than for bulk jobs: speech to text takes clips of up to 120 seconds with no streaming, text to speech returns at most 30 minutes of audio per request, and OCR stops at 1,000 pages a call.
As we read the billing rules, the free period our October piece reported for search and fetch calls made through DeepSeek Harness, Claude Code or Codex is not mentioned in the Search and Fetch guide we read on 13 November, which says such calls are billed per successful request.
DeepSeek, API change log entry of 13 November 2026
“Today we open these capabilities to developers: enable the corresponding parameters in the existing APIs to get conversation features such as web search, file input and URL reading, with the tool calls and context assembly handled by the service.”
Three Implications
FOR DEVELOPERS BUILDING ON DEEPSEEK The chat features arrive as parameters on the existing Messages, Responses and Chat Completions APIs, and the standalone APIs are priced per page, per hour of audio or per 10,000 characters on the pricing page.
FOR TEAMS BUDGETING VOICE FEATURES Both speech prices are labelled a “Limited-time price (50% off)”; the list prices are $0.116 per hour of audio and $0.45 per 10,000 characters. As we read it, the list prices are the safer basis for planning.
The framework opens: “The surface, not the model, governs autonomy. The agentic surface — the harness — is becoming where value pools in the entire AI stack.”
As we read it, DeepSeek is pricing the parts of its own harness, search, file handling, voice and safety, as services other developers can rent around its model. The framework does not discuss DeepSeek.
What Is Not Established
The pages we read give no usage figures, customers or revenue for the new APIs, and we did not find an end date for the speech discount in them. The worked examples are our arithmetic at list or discounted prices; they leave out model tokens, which are billed separately. We read DeepSeek’s documentation and GitHub only, not any statement on its social accounts, and we did not contact DeepSeek for this piece.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
DeepSeek has turned the search, file, voice and safety features of its chat app into priced API services, adding OCR at $0.225 per 1,000 pages and two speech APIs at half their list prices for now, alongside the $0.75 Search and Fetch prices set in October. The next things to watch are an end date for the speech discount and any usage figures from DeepSeek.
A note on sourcing. This piece is based on DeepSeek’s API documentation, read in full on 13 November 2026: the change log entry for that day, the Models & Pricing page and the OCR, speech-to-text, text-to-speech, Search and Fetch, file parsing, web tools and model safety guides, plus the GitHub record of the chat-demo repository. A post on X flagged the release; we did not rely on it. Worked examples and request counts are our arithmetic. We did not contact DeepSeek for this piece.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
