China AI Hub
API / deepseek
deepseek API
Provider: deepseek · Type: official
deepseek is an API by deepseek, served at https://api.deepseek.com for official models, exposing DeepSeek-V3.2, DeepSeek-V4-Pro, DeepSeek-V4.1-Flash.
| Key fact | Value |
|---|---|
| Endpoint | https://api.deepseek.com |
| Authentication | Bearer API key (created at platform.deepseek.com/api_keys) |
| Streaming | Yes |
| Function calling | Yes |
| Tool calling | Yes |
| Structured output | Yes |
| Vision | Yes |
| Audio | Unknown |
| Rate limits | Concurrency: deepseek-flash 2500; deepseek-v4-pro 500 (per official pricing page) |
| Regions | — |
| Cloud providers | DeepSeek Platform |
| Documentation | https://api-docs.deepseek.com/ |
Context limits
| Model | Input limit | Output limit |
|---|---|---|
| deepseek-v4-1-flash | 1,048,576 | 393,216 |
| deepseek-v4-pro | 1,048,576 | 393,216 |
What it is. The DeepSeek API is DeepSeek’s official, OpenAI-compatible platform serving DeepSeek-V4.1-Flash and V4-Pro, with an additional Anthropic-compatible base URL. Why it matters. It is the price-floor API of the Chinese market — V4.1-Flash lists $0.15/1M input — and the most compatibility-complete (OpenAI, Responses and Anthropic surfaces in one platform). Key characteristics. Two base URLs (api.deepseek.com and api.deepseek.com/anthropic); 1M-token context and 384K max output on both models; thinking/non-thinking modes, JSON output, tool calls, automatic KV caching. What a professional should know. Peak/off-peak pricing applies (peak is 2x the listed off-peak), concurrency limits are 2500 (flash) / 500 (v4-pro), and serving regions are not stated in the official docs.
The DeepSeek API is an OpenAI-compatible platform with two base URLs: https://api.deepseek.com
(OpenAI-compatible, also used by the native Responses API) and https://api.deepseek.com/anthropic
(Anthropic-compatible). Authentication is a Bearer API key; streaming is supported.
Current models are deepseek-flash (V4.1-Flash, multimodal) and deepseek-v4-pro, both with 1M-token
context and 384K max output, thinking and non-thinking modes, JSON output, tool calls, and automatic KV
context caching. Peak/off-peak pricing applies. FIM and Chat Prefix Completion are in beta
(non-thinking only). The serving regions are not stated in the official docs.
Why it matters
The DeepSeek API matters as the cost-and-compatibility reference point: its $0.15/1M flash input price is the lowest budget tier in the database, and its triple surface (OpenAI / Responses / Anthropic) minimizes integration friction for teams migrating from either ecosystem. The trade-off is operational opacity — no published serving regions and a peak/off-peak structure that doubles cost during peak windows. China AI Hub analysis indicates the API’s structural role is the low-cost default: it is the surface most likely to be chosen on price alone, with compatibility as the secondary moat and region transparency as the notable gap.
Labels used above: Official fact (from DeepSeek API docs and pricing page), Vendor-reported claim (pricing and concurrency figures published by DeepSeek), and China AI Hub analysis (our synthesis, always introduced as such).
Provider
Models served
Pricing
Sources
Data interpretation
| Field | Value | Evidence type |
|---|---|---|
| Endpoint | https://api.deepseek.com | Official |
| Authentication | Bearer API key (created at platform.deepseek.com/api_keys) | Official |
| Capabilities | Streaming, Function calling, Tool calling, Structured output, Vision | Official |
| Models served | 3 models in the database | China AI Hub analysis |
Evidence types: Official = vendor documentation or official pages. China AI Hub analysis = derived from the database itself. See the sourcing policy.
What is the deepseek API endpoint?
The deepseek API endpoint is https://api.deepseek.com.
How do you authenticate with the deepseek API?
Bearer API key (created at platform.deepseek.com/api_keys).
What are the context limits of the deepseek API?
deepseek-v4-1-flash: 1,048,576 in / 393,216 out; deepseek-v4-pro: 1,048,576 in / 393,216 out.
Does the deepseek API support function calling?
Yes — the deepseek API supports function calling.
Where does China AI Hub get its deepseek API data?
From 2 sources, last verified 2026-09-20.