DeepSeek

DeepSeek is a Chinese AI lab that sells API access to its own models and runs a free chat assistant. Its developer docs list two current API models. The first, called deepseek-flash, is served by DeepSeek-V4.1-Flash-1017 and supports image input. The second, deepseek-v4-pro, is served by DeepSeek-V4-Pro-0813 and handles text only. Both offer a context length of one million tokens, a maximum output of 384,000 tokens, and a thinking mode that is on by default and can be switched off.

Pricing is per million tokens and changes with the time of day. At peak rates, Flash costs $0.30 for input that misses the cache and $1.20 for output, while Pro costs $1.32 and $3.96. Off peak rates are half of those. DeepSeek defines peak hours as 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays, which is 9:00 PM to 12:00 AM and 2:00 AM to 6:00 AM ET during daylight time. Weekends and Chinese public holidays are off peak all day. Input that hits the cache costs far less, from $0.003 per million tokens on Flash off peak.

The API accepts both the OpenAI request format at api.deepseek.com and the Anthropic format at api.deepseek.com/anthropic, so many existing SDKs and coding tools work after changing the base URL. Features include JSON output, tool calls and the Responses API. Separate Search and Fetch APIs cost $0.75 per 1,000 successful requests. DeepSeek says calls made through tools such as Claude Code or Codex are free for now, with billing starting November 1, 2026 Beijing time.

Usage is prepaid: charges come out of a topped up balance, and any granted balance is used first. DeepSeek notes that prices may change and recommends checking its pricing page often. On intelsi.si we cover DeepSeek as a frontier lab and API provider; running its open weight models on your own hardware belongs to a different topic.