DeepSeek

DeepSeek is a large language model lab whose chat assistant became one of the most talked-about alternatives to ChatGPT. The company is based in China (its website carries Zhejiang ICP registration numbers), and it ships three things: a web chat and mobile app, a developer API, and a newer desktop agent called DeepSeek Harness, which is still in developer preview.

What you get in the chat

The web chat has two switches that matter: DeepThink, its reasoning mode, and Search, which lets the model pull in web results. In August 2026 DeepSeek added its larger V4-Pro model to the app and web through an "Expert Mode." On 17 October 2026 the Flash model was updated to V4.1-Flash-1017, which DeepSeek says handles project development work faster and at lower cost.

The API is where the pricing lives

DeepSeek doesn't advertise a subscription for the chat or app. Its pricing page covers only the API, and that pricing is unusually low. Two models are offered, both with a 1M-token context window and up to 384K tokens of output:

• deepseek-flash: $0.30 per million input tokens (cache miss) and $1.20 per million output tokens at peak; half that off-peak. Flash also handles images.
• deepseek-v4-pro: $1.32 input and $3.96 output per million tokens at peak, $0.66 and $1.98 off-peak. No vision support.

Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays; everything else, including weekends, bills at the off-peak rate. The API speaks both the OpenAI and Anthropic request formats, so tools like Claude Code, Codex or OpenCode can point at it directly. New Search and Fetch endpoints cost $0.75 per 1,000 successful requests, and DeepSeek says server-side search calls made through those coding tools become billable from 1 November 2026. Usage is deducted from a prepaid balance.

Prices checked October 2026 on the official DeepSeek API pricing page.

Good to know

Pros: very low API prices, especially off-peak; huge context window; drop-in compatibility with OpenAI and Anthropic SDKs; reasoning and web search built into the chat, with no paid plan advertised for it.

Cons: the two-tier peak/off-peak pricing takes some planning; there's no published plan or usage limit for the consumer app; and companies with strict data-residency rules will want to read DeepSeek's privacy terms before sending it sensitive material.

It suits developers who push a lot of tokens through a model and watch costs, and anyone who wants a capable chatbot with a reasoning mode and no subscription.

Alternatives

If you'd rather stay with a US-based provider, look at ChatGPT or Claude. Kimi is another Chinese model with a long context window, and Grok is worth a look if you want live web answers.