What happened
In mid-July 2026, DeepSeek moved its V4 models from preview to full release. The line has two models. V4-Pro is very large, at 1.6 trillion parameters, the settings an AI learns in training. V4-Flash is smaller, at 284 billion, and can read up to 1 million tokens at once. Tokens are the small chunks of text an AI reads and writes. One million is roughly a very long book in a single prompt.
The context
V4 is DeepSeek's push toward the frontier. Its large memory and low cost are the main selling points.
What this means for you
👤EverydayApps built on V4 can handle longer chats and files without losing track.
💼At workYou can feed in big documents or code and ask questions across all of it.
🏢BusinessA capable, low-cost model gives buyers a strong value option to weigh.
🌍The worldOpen Chinese models keep closing the gap with top US systems.
Ledger entry: Release · Materiality 3 · T3 Reported → bounded ▲ for DeepSeek