LLM Evaluation

DeepSeek V4.1 Flash: Smaller KV Cache, Cheaper Writing Drafts

A larger backbone, a smaller global cache, and drafts averaging less than a cent at default effort in our writing test. Here is what the numbers mean.