DeepSeek V4-Pro
1.6-trillion-parameter preview that claims to 'close the gap' with frontier closed models — April 2026
Shipped by DeepSeek · · class: open-weight-llm
On April 24, 2026, DeepSeek released a preview of its V4 series. The flagship V4-Pro is a 1.6-trillion-parameter mixture-of-experts model; V4-Flash is a 284-billion-parameter companion. Both feature a 1-million-token context window and are released under the MIT License, continuing DeepSeek’s open-weight discipline.
What was claimed at release: V4 has “almost closed the gap” with current leading closed-frontier models on the benchmarks DeepSeek selected. Architectural changes deliver materially better tokens-per-second-per-dollar than V3.2 for comparable quality.
What’s being observed: the claim is partly testable — independent benchmark reproduction is in progress through May 2026. The bigger discourse-level effect is the renewed pricing pressure on closed-frontier labs, with several inference providers cutting prices within a week of the V4 preview. Whether the “closed the gap” framing survives third-party evaluation is a tracked claim (see /claims/).