The efficiency-first member of the V4 line: MoE, ~284B total and only ~13B active per token, context up to 1M. Tuned for fast, cheap inference — everyday chat and light tasks.
The efficiency-first member of the V4 line: MoE, ~284B total and only ~13B active per token, context up to 1M. Tuned for fast, cheap inference — everyday chat and light tasks.
The efficiency-first member of the V4 line: MoE, ~284B total and only ~13B active per token, context up to 1M. Tuned for fast, cheap inference — everyday chat and light tasks.
The efficiency-first member of the V4 line: MoE, ~284B total and only ~13B active per token, context up to 1M. Tuned for fast, cheap inference — everyday chat and light tasks.
DeepSeek V4 Flash 0731 is a sparse mixture-of-experts model from DeepSeek, with 13B active parameters out of 284B total. This re-post-trained revision is suited for coding, reasoning, and agent workflows.