2026-10-12 · DGX Spark Class
A model scorecard round on DGX Spark Class · NVIDIA GB10 · 122 GB: 2 models across 4 task sets. Every round is listed in the model scorecard, with its results side by side in one table.
DGX Spark Class · NVIDIA GB10 · 122 GB · llama-cpp · 3 trials per task · 2026-10-12 · gezel da2ed808 · catalog 0.1.95
Measured after the October 2026 fixes to our tests.
Memory isn't shown for this round. This machine's graphics chip shares the computer's memory in a way our measurement can't see, so the number would read far below what the model really uses.
General capability
| Model | Size | Tasks passed | Performance | Context |
|---|---|---|---|---|
| qwen3.8-27b-q4 (kv: f16) | 27B | 33/33 (100%) | 30.1 tok/s output 699 tok/s prefill | 256K |
| gemma4-31b-q4 (kv: f16) | 30.7B | 32/33 (97%) | 33.5 tok/s output 754 tok/s prefill | 81K |
Tasks in this set: 11.
Office and knowledge work
| Model | Size | Tasks passed | Performance | Context |
|---|---|---|---|---|
| qwen3.8-27b-q4 (kv: f16) | 27B | 37/39 (95%) | 30.1 tok/s output 699 tok/s prefill | 256K |
| gemma4-31b-q4 (kv: f16) | 30.7B | 36/39 (92%) | 33.5 tok/s output 754 tok/s prefill | 81K |
Tasks in this set: 13.
Engineering work
| Model | Size | Tasks passed | Performance | Context |
|---|---|---|---|---|
| qwen3.8-27b-q4 (kv: f16) | 27B | 29/30 (97%) | 30.1 tok/s output 699 tok/s prefill | 256K |
| gemma4-31b-q4 (kv: f16) | 30.7B | 26/30 (87%) | 33.5 tok/s output 754 tok/s prefill | 81K |
Tasks in this set: 10.
Complex workflows
| Model | Size | Tasks passed | Performance | Context |
|---|---|---|---|---|
| gemma4-31b-q4 (kv: f16) | 30.7B | 20/25 (80%) | 34 tok/s output 757 tok/s prefill | 81K |
Not published — some tasks could not be measured on this round: qwen3.8-27b-q4 (1 task(s) unmeasured)
Tasks in this set: 9.