DeepSeek's v4.1-Flash is described in Latent Space's AI news roundup as a major architectural release. The model is listed as 763B-P8B-D16B and uses a causal encoder-decoder design with vision capabilities, according to the source.
The newsletter frames the release as "the Return of the Whale," and the author agrees with Sebastian that the version number undersells the change: "this should have been DeepSeek v5." The emphasis is on the novelty of the architecture rather than on incremental improvements.
Since there is only one source, there are no differing accounts to weigh. The excerpt does not include benchmark results or an explanation of the parameter notation, so the practical performance of the model remains an open question.