DeepSeek-V4-Flash: an open model with 284 billion parameters that activates only 13 at a time
On July 31, 2026, DeepSeek released the official version of V4-Flash, an open MoE model with 284 billion parameters under the MIT license and a one million token context. The smaller model outperforms the larger V4-Pro on agent tasks. I ran it on my Mac, streaming from the SSD.
Preview
This is a short excerpt.
The full version, with its sources, and the podcast are for subscribers. The account is free.