DeepSeek
6.4/10DeepSeek is a Chinese AI lab that offers a chat app and a very low-cost LLM API for text, reasoning, and coding, aimed mainly at developers.
Data last updated November 2026 · Our methodology
Overview
DeepSeek is a Hangzhou-based AI company, backed by the quant firm High-Flyer. It offers a web chat at chat.deepseek.com, a mobile app, a desktop product called Harness, and a developer API platform. The API serves two current models. deepseek-flash (V4.1-Flash) supports vision input. deepseek-v4-pro is the stronger, more expensive model. Both have a 1M-token context window, up to 384K output tokens, thinking and non-thinking modes, JSON output, tool calling, and fill-in-the-middle completion. Both are compatible with the OpenAI-style Responses API and the Anthropic API formats, so existing SDKs and agent tooling can switch to DeepSeek with little effort. The API recently added web search and fetch, and speech APIs for text-to-speech and speech recognition. Pricing is pay-as-you-go per million tokens, with cheaper rates for cached input, and off-peak hours cost half the peak rate. That makes it one of the cheapest frontier-class APIs, which suits startups, developers, and high-volume batch workloads. The main limitation is trust and compliance. The hosted service processes data on servers in China. Reviewers report no SOC 2, SSO, or audit logs. In October 2026, US federal civilian agencies were ordered to stop using DeepSeek, and Australia has banned it on government systems. Regulated and enterprise buyers often run the models through third-party hosts or self-hosted deployments instead of DeepSeek's own API.
Our Rating
Who DeepSeek is best for
Best software for DevelopersGood fit
Its OpenAI- and Anthropic-compatible API, tool calling, and very low token prices make it a natural fit for builders.
Pay-as-you-go pricing with off-peak discounts keeps LLM costs very low for early-stage products.
Less suited
Data is hosted in China and reviewers report no SOC 2, SSO, or audit logs.
Security reviewers advise against using it with PHI because no BAA is available.
US federal civilian agencies and Australian government systems have banned it.
Pros & Cons
Pros
- Token prices are among the lowest of any frontier-class API, and off-peak hours halve them again.
- The API works with the OpenAI Responses and Anthropic formats, so existing SDKs and agents can switch over easily.
- Both models have a 1M-token context window and up to 384K output tokens.
- Cached input is billed at a small fraction of the normal input rate, which suits repeated-prompt workloads.
- Reviewers rate its reasoning and coding output as strong for the price.
Cons
- The hosted API and app process data on servers in China, so security teams often refuse to approve it.
- US federal civilian agencies have been ordered to stop using it, and Australia has banned it on government systems.
- Reviewers report no SOC 2, SSO, audit logs, or BAA.
- It filters or refuses politically sensitive China-related topics.
- Reviewers report slow responses at peak times and occasional outages.
Key Features
Not included
- Image Generation
- Enterprise Controls
DeepSeek Pricing
API pay-as-you-go
- 1M-token context window
- Up to 384K output tokens
- Usage-based billing per 1M tokens
- Off-peak rates are 50% of peak rates
- Concurrency limit of 2,500 on Flash and 500 on Pro
Frequently Asked Questions
Reviews describe the web chat as free. API use is billed per million tokens.
Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays, excluding Chinese public holidays. Outside those hours, rates are half the peak price.
Yes. The API supports the Responses API and Anthropic API formats, along with tool calls and JSON output.
Use caution. Data on the hosted service goes to China. Security reviewers advise against using it for PHI or PCI data, and US federal civilian agencies have banned it.
Deepseek-flash accepts vision input. Text-to-speech and speech recognition APIs are also offered.
Similar Tools
See all alternativesOpen-source Python and TypeScript framework from OpenAI for building multi-agent applications with tools, handoffs, guardrails, sessions, and built-in tracing.
Open-source Python framework from deepset for building production LLM applications, including RAG pipelines, agents and semantic search, with optional paid enterprise support.
Google's family of generative AI models and assistants, sold to businesses through Gemini Enterprise, Workspace plans and a pay-per-token developer API.
Ranked near DeepSeek in AI Software
See DeepSeek in the listDeepSeek is #358 in AI Software by our score.
