DeepSeek

DeepSeek is a Chinese AI research company that builds large language models and ships them through a consumer assistant and a developer API. The assistant runs on the web, on mobile apps and on a desktop client, with a deep thinking mode for extended reasoning and a search mode that grounds answers in web results. The company also releases model weights publicly, which is how its work reached most practitioners. The developer platform is the part most relevant to builders. Two models are offered: a fast general model with native multimodal image understanding, and a larger reasoning focused model. Both carry a context window of one million tokens and a maximum output of 384,000 tokens, both support JSON output, tool calls, prefix completion and fill in the middle completion in beta, and both are reachable through endpoints compatible with the OpenAI and Anthropic request formats, which makes switching an existing integration largely a matter of changing the base URL. Concurrency limits differ sharply between the two tiers. API pricing is published per million tokens in Chinese yuan and varies by time of day, with off peak rates set at half the peak rate. On the fast model, cached input is 0.02 yuan off peak and 0.04 yuan at peak, uncached input is 1 yuan and 2 yuan, and output is 4 yuan and 8 yuan. On the larger model the same three lines are 0.15 and 0.30 yuan, 4.5 and 9.0 yuan, and 13.5 and 27.0 yuan. Usage is drawn from a prepaid balance, and no subscription price is published for the consumer assistant.