DeepSeek-V4-Flash

DeepSeek-V4-Flash is a large language model in DeepSeek's V4 series, positioned as a lighter-weight counterpart to the flagship V4 Pro. As described in a 2026-08-15 十字路口Crossing piece, it was priced at $0.28 per million output tokens before DeepSeek switched to peak-valley pricing on August 17, 2026, when Flash's rates moved to $0.66 off-peak and $1.32 peak . In benchmark comparisons around that period, the model was noted for punching above its weight: 葬AI observed that Flash delivered performance comparable to GLM-5.2—a model roughly triple its parameter size—suggesting strong efficiency in post-training optimization . The model is typically evaluated alongside DeepSeek's official Harness (DSH), where its high cache hit rates (reportedly near 99% in one test) help keep total task costs low even as per-token pricing rises .