← Leaderboard

qwen3-coder-next

MoE · 3B active / 80B total vLLM AgentsCode

Throughput

43 22 0 38.7 4k 38.4 50k 37.5 100k tok/s

Decode tok/s after a fixed context fill (PBM) — same agent-style workload as bench v2.

Recipe

Profile
qwen-qwen3-coder-next-eugr
Engine
vLLM
Context
256k · fp8 KV
Served as
qwen3-coder-next

Run on your Spark

single node shell
spark inference up qwen-qwen3-coder-next-eugr

Why we run it

Golden fleet target — auto-scaffolded from recipe qwen-qwen3-coder-next-eugr.

Bench notes

PBM 4k @ 38.7 tok/s — perfbench-metrics — profile=qwen-qwen3-coder-next-eugr

Measurement history

Context ladder

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

Context KV Throughput
@ 256k peak golden 38.7t/s

Benchmark runs

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

Date Profile Method Avg Session t/s Range Fill Tool
2026-06-28latest qwen-qwen3-coder-next-eugr bench-agent-v2v2.0 38.7t/s ~50,000 fail
golden 256k/fp8 @ 38.7 tok/s — fill~50000 — bench-agent-v2 — tool_ok=False
Benchmarked 2026-07-10
SparkBench · GB10 · single node