← Leaderboard
google/gemma-4-26b-a4b-it

Gemma-4-26B-A4B-IT

vLLM MoE · 4B active / 26B total GeneralMultimodalReasoning
Throughput
23.6t/s@ 8k
Engine
vLLM
Parameters
4B active / 26B total
Released
2026-03-11
Benchmarked
2026-06-27

Context ladder

Throughput at each benched context window.

Context KV Throughput
@ 8k peak 23.6t/s
@ 256k golden fp8 22.2t/s

Golden profile

google-gemma-4-26b-a4b-it-eugr

Capabilities

generalmoemultimodalvisionagenticcodingvllmapache-2.0

Why we run it

Popular MoE pick (~3.8B active / 26B total) — same efficiency class as Qwen3.6 MoE. 256K context, text+image. Heavier than 12B but fast inference per token.

Bench notes

golden 256k/fp8 @ 22.2 tok/s — fill~50000 — bench-agent-v2 — tool_ok=False

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