← Leaderboard

DeepSeek V4 Flash (DwarfStar)

MoE · 13B active / 180B total ds4 AgentsCodeReasoning

Recipe

Profile
antirez-deepseek-v4-flash-ds4
Engine
ds4
Context
32k · auto KV
Served as
deepseek-v4-flash

Run on your Spark

single node shell
spark inference up antirez-deepseek-v4-flash-ds4

Why we run it

Production DSv4-Flash on GB10 via ds4 engine — 0731 + DSpark.

Bench notes

Production 0731+DSpark on Entrpi/ds4 v0.5.0. bench-v2 19.5 tok/s @ ~14k fill (tool_ok=True). Short-decode mean ~36.6 (math 45, code 39, prose 22).

Measurement history

Context ladder

Older bench-v2 / golden cells at each benched context window (single measurement).

Context KV Throughput
@ 32k peak 19.5t/s

Benchmark runs

Recorded inference benchmark sessions for this model's profile (single run).

Date Profile Method Avg Session t/s Range Fill Tool
2026-08-04latest antirez-deepseek-v4-flash-ds4 bench-agent-v2v2.0 19.5t/s 19.5 · 19.5 ~14,745 ok
bench-v2 avg 19.5 decode tok/s (2 sessions, ~14k ctx fill, tool_ok=True)
Benchmarked 2026-08-04
SparkBench · GB10 · single node