DeepSeek V4 Pro 🧠 Model Open source

What it is

DeepSeek is a Chinese AI lab that publishes open-weight language models. Its current generation, DeepSeek V4, launched in April 2026 in two sizes: V4 Pro (1.6T parameters, 49B active) and V4 Flash (284B, 13B active). DeepSeek V4 Pro left preview and became generally available on the API and in the chat app on 13 August 2026. Both models have a 1M-token context window.

Key features

• 1M-token context and up to 384K output tokens per response
• Reasoning effort setting (low, high, max) so simple rewrites stay cheap
• OpenAI-compatible API, including the Responses API
• Flash supports image input, JSON output and tool calls; V4 Pro is text-only
• Weights on Hugging Face under MIT, so you can fine-tune on your own style

Where you can use it

In DeepSeek's free chat app (web and mobile, with Expert Mode on V4 Pro) and the DeepSeek API. The weights run on vLLM or SGLang and through many hosting providers. For content automation, n8n has a DeepSeek Chat Model node.

Pricing and rights

Since 16 August 2026 the API uses off-peak and peak rates; peak (weekday mornings UTC) costs double. Off-peak list prices per million tokens: Flash input from $0.003 on cache hits and $0.60 output; V4 Pro input from $0.022 on cache hits and $1.98 output. The MIT licence allows commercial use of the weights and outputs. The chat app is free, so there is no paid consumer plan.

Who it is for

Budget-conscious publishers running high-volume drafting, rewriting or translation, and teams who want open weights they can host themselves.

Verdict

DeepSeek V4 offers frontier-class context at a fraction of Western API prices, with a truly permissive licence. The flagship is text-only, peak pricing complicates budgeting, and some clients will object to sending data to a China-based service.