Qwen3.8-Flash-Next INT4-AR
Editor's pick — the recipe we'd run first on a Spark right now.
Throughput
Decode tok/s after a fixed context fill (PBM) — same agent-style workload as bench v2.
Published evals
Vendor-card figures for qwen/qwen3.8-flash-next — not this pack. Not measured on this Spark.
- Terminal-Bench 2.1
- 82.9% NVIDIA NVFP4 model card ↗ 2026-08-31 NVIDIA NVFP4 column; temp=1.0, top_p=0.95, reasoning_effort=xhigh. FP8 baseline on the same card is 83.3. Not the Mia pack, not measured on Sparky.
- SWE-bench Pro
- 62.5% Qwen model card ↗ 2026-08-26 Claude Code harness; refined SWE-bench Pro set (vendor). DeepSWE 1.1 is 58.7 on the same card (not a site column).
- LiveCodeBench
- 91.9% Qwen model card ↗ 2026-08-26 LiveCodeBench v6 (vendor).
- GPQA Diamond
- 91.7% Qwen model card ↗ 2026-08-26
Independent board on Artificial Analysis ↗ — we do not copy their scores.
Recipe
- Profile
- azampatti-qwen3.8-flash-next-int4-autoround-tensorfold
- Engine
- TensorFold
- Context
- 256k · fp8 KV
- Served as
- Qwen3.8-Flash-Next
- Draft
- MTP
Run on your Spark
spark inference up azampatti-qwen3.8-flash-next-int4-autoround-tensorfold
Why we run it
Qwen3.8-Flash-Next on one Spark via azampatti INT4-AutoRound (~122 GiB, top-5 experts, ~4.8B active) on MiaAI TensorFold. Native 262k, FP8 KV, vision and MTP drafts on, PARALLEL 8. PBM 50.2/26.6/15.4 tok/s. This is the Flash-Next ladder row.
Bench notes
PBM 4k @ 50.2 / 50k @ 26.6 / 100k @ 15.4 tok/s — perfbench-metrics — native 262k TensorFold INT4-AR vision on, PARALLEL 8 — profile=azampatti-qwen3.8-flash-next-int4-autoround-tensorfold