← Leaderboard
unsloth/qwen3-coder-30b-a3b-instruct

Qwen3-Coder-30B-A3B-Instruct

llama.cpp MoE · 3B active / 30B total AgentsCode
Throughput
70.9t/s@ 32k
Engine
llama.cpp
Parameters
3B active / 30B total
Released
2025-07-31
Benchmarked
2026-06-27

Context ladder

Throughput at each benched context window (single measurement).

Context KV Throughput
@ 32k peak golden 70.9t/s

Golden profile

unsloth-qwen3-coder-30b-a3b-instruct-llama

Capabilities

codingagenticmoellamacppgguf

Why we run it

DGX Spark forum pick for agentic coding. MoE coder fits 128GB; Q4 + Q5 GGUF for llama.cpp bake-off path.

Bench notes

golden ?/? @ 29.2 tok/s — fill~14745 — bench-agent-v2 — tool_ok=True

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