← Leaderboard
google/diffusiongemma-26b-a4b-it

Diffusiongemma 26B A4B It

vLLM MoE · 4B active / 26B total MultimodalReasoning
Throughput
70.8t/s@ 16k
Engine
vLLM
Parameters
4B active / 26B total
Released
2026-06-09
Benchmarked
2026-06-27

Context ladder

Throughput at each benched context window.

Context KV Throughput
@ 16k peak 70.8t/s
@ 256k golden fp8 39.3t/s

Golden profile

google-diffusiongemma-26b-a4b-it-eugr

Capabilities

explorervllm

Why we run it

HF Explorer download 2026-06-22

Bench notes

golden 256k/fp8 @ 39.3 tok/s — fill~50000 — bench-agent-v2 — tool_ok=False

Benchmarked 2026-06-27
SparkBench · GB10 · single node