What big models actually run at on DGX Spark / GB10 hardware, from one node to eight — measured as tokens delivered to the client, not server-log numbers. Every figure follows the methodology. If a number here surprises you, that's the point: most published Spark numbers measure the wrong thing.
| model | size | nodes | quant | decode (code) | detail |
|---|---|---|---|---|---|
| Kimi K2-class | ~1T MoE | 8 × GB10 · 100GbE | nvfp4 | 26–28 tok/s | full entry → |
| Qwen3.6-Next | 35B-A3B MoE | 1 × GB10 | bf16 | 45.2 tok/s | full entry → |
| GLM-5-class | 753B MoE | 8 × GB10 · 100GbE | mixed 4-bit | 16.5 tok/s (tuning in progress) | full entry → |
| MiniMax M3-class | ~400B MoE + vision | 8 × GB10 · 100GbE | w4a16 | 34.5 tok/s | full entry → |
All entries are single-stream decode on code workloads unless labeled otherwise; each full entry breaks out prose, prefill, aggregate, and long-context behavior where measured. Numbers are from our own 8× GB10 cluster on an ordinary 100 Gbps switch — the config most owners actually have, not a lab exotic.
Nobody publishes reproducible multi-node Spark numbers for models this size — the biggest public cluster results are video demos without configs. Every entry here is the end state of a measured tuning journey: what the model did out of the box, what class of problem was in the way, and what it runs at now. The journeys are described; the exact fixes are what the diagnostic delivers for your cluster.
We add entries as we tune them. If you need a number for a specific model on Spark hardware before you buy or deploy — that's a service we sell, with the harness output included.
Ask for a benchmark →