DeepSeek
Frontier AI models at commodity prices: DeepSeek's chat, API, and open weights for building.
DeepSeek is an AI lab offering a free chat app, a low-cost API ($0.15/M input), and MIT-licensed open model weights, including the 1M-context V4.1-Flash family.
DeepSeek homepage with chat interface and V4.1-Flash-1017 announcement
Overview
What is DeepSeek?
Frontier AI models at commodity prices: DeepSeek's chat, API, and open weights for building.
DeepSeek is a Hangzhou-based AI lab that builds and serves its own large language models, known for pricing frontier-level capability at a fraction of typical API rates. Its flagship products are the free DeepSeek chat app (web, iOS, and Android), a pay-per-token API platform with OpenAI- and Anthropic-compatible endpoints, and a family of open model weights released under the MIT license.
The October 2026 V4.1-Flash-1017 checkpoint brought agentic coding benchmarks to 77.0 on DeepSWE v1.1 with a 1M-token context window, API pricing of $0.15 per million input tokens and $0.60 per million output tokens off-peak, and cache reads at $0.003/M. Alongside the models, DeepSeek opened its in-house Search and Fetch APIs ($0.75 per 1,000 successful calls), giving developers a model plus a first-party web layer behind one API key.
For teams, the practical appeal is unit economics: long-context agent workloads with heavy cache reuse run at a small fraction of the cost of comparable frontier APIs, and the MIT-licensed weights allow self-hosting and fine-tuning without usage tiers.
Platforms and languages
DeepSeek Availability
Platforms
Languages
Capabilities
DeepSeek Key Features
Free chat app
Conversational AI on web, iOS, and Android with deep-thinking and smart-search modes.
Low-cost API
OpenAI- and Anthropic-compatible endpoints; Flash tier at $0.15/M input and $0.60/M output off-peak.
1M-token context
Long-context models with very large output budgets for agent and RAG workloads.
MIT open weights
Checkpoints such as V4.1-Flash-1017 on Hugging Face and ModelScope, free for commercial use and fine-tuning.
Search and Fetch APIs
First-party web search (up to 50 ranked results) and page-to-Markdown fetch behind the same API key.
Controllable reasoning
Thinking and non-thinking modes with adjustable reasoning effort for complex tasks.
Plans and access
DeepSeek Pricing
Freemium
Chat app is free. API is pay-per-token: $0.15/M input and $0.60/M output off-peak; cached input $0.003/M.
Common questions
DeepSeek FAQs
Is DeepSeek free to use?
Yes. The DeepSeek chat app (web, iOS, Android) is free to use with generous limits. API access is billed per token, with the Flash tier at $0.15 per million input tokens and $0.60 per million output tokens off-peak.
How do I use DeepSeek's API with the latest model?
Call the API with the model name `deepseek-flash`. Endpoints are OpenAI- and Anthropic-compatible, so most existing integrations work by changing the base URL and key. Recommended sampling for the 1017 checkpoint is temperature 1.0 / top_p 1.0.
Are DeepSeek's model weights open source?
Yes. Checkpoints such as DeepSeek-V4.1-Flash-1017 are published on Hugging Face and ModelScope under the MIT license, so commercial use, fine-tuning, and redistribution are permitted.
What are DeepSeek's context limits and API prices?
The API offers a 1M-token context window with very large output budgets. Off-peak pricing is $0.15/M uncached input ($0.003/M for cache reads) and $0.60/M output; peak-hour rates are double. See api-docs.deepseek.com for the current price list.
Reviews
See what the community thinks and share your experience.
—
Based on 0 ratings
Rating distribution
Community reviews
No reviews
No written reviews yet.
