DeepSeek V4 Flash 284B (13B active)

on 2× NVIDIA RTX 5090 · llama.cpp · 384,000 ctx

Sep 7, 2026

Throughput

30.2 t/s gen · 64.0 tokens/s prompt processing (prefill)Prompt processing measures input prefill speed; it does not tell you the time to first token.

Quant

Q8_K_XL

KV cache

bf16

System RAM

96 GB

Use cases

agentictool-usereasoning

Summary

User benchmarks Q8_K_XL against IQ3_XXS on DeepSeek-V4-Flash-0731-UD. Q8_K_XL offloads about 40 GB to system RAM. IQ3_XXS is 2.2x faster at decode and 2.4x faster at prefill, with equal correctness. Q8_K_XL has more empty-answer failures, 8/125 against 5/125. A reasoning budget of 4096 eliminates empty answers, and disabling reasoning degrades trap handling. Q8_K_XL is slower but more reliable in some tasks.

View source ↗