DeepSeek has been the open-weight world's most-watched wildcard since R1 rattled the market in early 2025. On 24 April 2026 it finally shipped the long-awaited next step: DeepSeek-V4 Preview.

Two models, both big

The preview comprises two Mixture-of-Experts models:

• DeepSeek-V4-Pro — 1.6 trillion total parameters, 49B activated per token.
• DeepSeek-V4-Flash — 284 billion total parameters, 13B activated.

Both support a 1-million-token context window. The MoE design is the usual efficiency play: enormous total capacity, but only a fraction of parameters fire on any given token, keeping inference cost far below what the headline parameter count implies. That's been DeepSeek's signature — frontier-adjacent capability at a cost structure that undercuts Western labs.

The pitch

DeepSeek frames V4 around stronger agentic capabilities and top-tier reasoning — the same two axes everyone is competing on. The interesting part is positioning: V4 lands as a preview, available via DeepSeek's API as deepseek-v4-pro and deepseek-v4-flash, with weights for the family circulating in the open-weight ecosystem. For teams that need to run a capable model on their own infrastructure — for cost, data-residency or sovereignty reasons — a trillion-parameter-class MoE with a million-token window is a serious option.

The R2 question

A necessary note on what V4 is not. For over a year, the AI commentariat has anticipated a DeepSeek R2 — a successor to the R1 reasoning model that made DeepSeek a household name. As of the checks underlying this piece, there is still no officially released R2: it does not appear in DeepSeek's current API model list, and no official release date has been confirmed. If you see a confident 'R2 is here' claim, treat it sceptically until DeepSeek lists it officially. The real, shipped artifact is V4 Preview — and that's plenty.

Why it matters

DeepSeek's role in the ecosystem is to keep the open tier honest. Every time it ships a capable model at aggressive cost, it compresses the premium that closed labs can charge and pulls the price floor down for everyone. V4 Preview, with its scale and context length, does exactly that — it keeps the downloadable frontier within touching distance of the closed one, which is precisely why the rest of the field watches DeepSeek so closely.