DeepSeek

DeepSeek is a conversational AI assistant from the Chinese lab of the same name. It offers a web chat interface at chat.deepseek.com alongside a metered API that developers can build on. It handles general chat, coding, reasoning, and long-document work.

The API currently serves two model families: deepseek-flash, now on the V4.1-Flash architecture released September 10, 2026 with a native multimodal design, and deepseek-v4-pro, which reached general availability on August 13, 2026. Older model names such as deepseek-chat and deepseek-reasoner no longer appear on the pricing page; legacy aliases deepseek-v4-flash and deepseek-v4-flash-vision-exp are still accepted but now route to V4.1-Flash. DeepSeek has iterated on its model lineup rapidly over the past few months, so names and pricing have changed more than once in a short period.

Both models offer a 1M-token context window with output up to 384K tokens, and a default thinking mode with selectable low, high, and max reasoning effort. Vision input is supported on deepseek-flash but not on deepseek-v4-pro. DeepSeek documents agent integrations including Claude Code, Codex, OpenCode, and its own DeepSeek Harness.

API pricing is billed per 1 million tokens and varies by peak and off-peak UTC hours, with off-peak priced at half the peak rate. Peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, excluding Chinese public holidays. For deepseek-flash: input on a cache miss is $0.15 off-peak / $0.30 peak, input on a cache hit is $0.003 off-peak / $0.006 peak, and output is $0.60 off-peak / $1.20 peak. For deepseek-v4-pro: input on a cache miss is $0.66 off-peak / $1.32 peak, input on a cache hit is $0.022 off-peak / $0.044 peak, and output is $1.98 off-peak / $3.96 peak. The pricing page also references a 'granted balance' used before any topped-up balance, but does not explain how or when it is granted. Pricing for the web chat interface could not be confirmed on a primary page at time of writing.

Limitations to keep in mind: whether the web chat remains free could not be re-verified today, the 'granted balance' mechanism is unclear, concurrency is capped at 2,500 requests for deepseek-flash and 500 for deepseek-v4-pro, vision is Flash-only, the model lineup has changed several times in recent months, and as a China-based service it may raise data-residency or compliance questions for some organizations.

DeepSeek suits developers who want a low-cost, long-context API for coding, reasoning, or high-volume tasks, and who are comfortable tracking peak/off-peak pricing and occasional model migrations.