DeepSeek capabilities

Extended chain-of-thought reasoning exposed through the DeepSeek API. When thinking mode is enabled, the model produces an explicit reasoning trace before its final answer to improve accuracy on math, logic, coding, and multi-step problems. The reasoning trace is returned to the developer in a dedicated reasoning_content field, separate from the final answer in content, so applications can display or discard the chain-of-thought as needed.

The DeepSeek API automatically caches repeated prompt prefixes on disk, dramatically reducing cost and first-token latency for workloads that reuse the same system prompt or long context across multiple requests.

DeepSeek persists conversation history across sessions, accessible in the sidebar. A chat history search feature allows users to find past conversations by content.

DeepSeek chat offers access to its frontier general-purpose models — currently DeepSeek-V4-Flash and DeepSeek-V4-Pro — for conversational AI, coding, math, and long-form reasoning.

All user data — including chat content, uploaded files, account details, and device signals — is stored on servers located in the People's Republic of China, subject to Chinese law, which may require disclosure to authorities without independent court oversight.

A togglable reasoning mode (the DeepThink button) that shows explicit chain-of-thought steps before delivering its final answer. Historically powered by DeepSeek-R1; on the current V4 app it activates the model's Thinking mode. Suited for complex math, logic, code debugging, and multi-step analysis.

A REST API exposing DeepSeek's frontier models to developers, compatible with the OpenAI and Anthropic SDK interfaces, supporting chat completions, tool calls, JSON output, FIM completion, and streaming.

Users can upload documents (PDF, DOCX, TXT, CSV, XLSX) and ask questions about their contents. DeepSeek extracts and processes text from the uploaded files to answer queries, summarize, or analyze.

The DeepSeek web and mobile chat interface is entirely free to use with no paid subscription tiers, no ads, and no per-message charges for end users.

Users can upload images (JPG, PNG, WebP) and DeepSeek will extract text via OCR and answer questions about the image content, supporting both printed and handwritten text.

DeepSeek publishes full model weights for its frontier models under the MIT license, allowing self-hosting, fine-tuning, distillation, and commercial use without restrictions.

An opt-in search toggle that queries the live web and integrates retrieved results into the model's response, providing up-to-date information beyond the model's training cutoff.

Official DeepSeek apps for iPhone/iPad and Android devices, providing full access to the chat assistant including DeepThink mode, web search, and file upload, synchronized with the web account.