DeepSeek V4 Flash 0731 (Scaleway)

DeepSeek-V4-Flash-0731 is the official release of the DeepSeek-V4-Flash variant, superseding the April 2026 preview and entering public API beta on July 31, 2026, per the OpenLLMStack model directory page. The checkpoint keeps the same architecture as the preview — a 284B-parameter Mixture-of-Experts with 13B active pa Attention interleaves Compressed Sparse Attention (CSA) with Heavily Compressed Attention (HCA), holding a 1,048,576-token context window without dense-attention memory cost, with vocab size 129,280 and FP4 + FP8 mixed precision. Two checkpoint-specific features stand out: a DSpark speculative decoding module attached