D
DeepSeek: deepseek-v4-flash-0731
deepseek-v4-flash-0731
1.0M contexte384K sortieRaisonnementOutilsOutils parallèlesCacheStructuré
Publié Jul 31, 2026Mis à jour Jul 31, 2026
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.
Tokenizer DeepSeekQuantification fp8