Coding with AI / GB10 (DGX Spark / GX10)

GB10 (DGX Spark / GX10)

Memory: 128 GB. Memory bandwidth: 273 GB/s. Generation speed is bound by the chip's bandwidth; memory decides which models fit.

How to run

Qwen3.6-35B-A3B smart

• Score: 51.5% (Terminal-Bench 2), source
• Speed: 55 tok/s (estimate: bandwidth-bound approximation (official llama.cpp GB10 bench: Qwen3-Coder-30B-A3B Q8_0 tg 61.06, близкая A3B MoE; Q4 35B чуть тяжелее по активным весам))
• Conditions: engine llama.cpp, quant Q4_K_M, context 128
• Min memory: 24 GB
• Command: llama-server -m qwen3.6-35b-a3b-Q4_K_M.gguf

Every number above is either a measurement with a source or a labeled estimate with a method. Unlabeled estimates do not exist on this site.

Ceiling timeline

The strongest model fitting GB10 (DGX Spark / GX10) (128 GB) at its release date, by SWE-bench Verified.

1. 2025-08-05 gpt-oss-20b 60.7%
2. 2025-08-05 gpt-oss-120b 62.4%
3. 2025-12-09 Devstral 2 72.2%
4. 2026-04-01 Qwen3.6-35B-A3B 73.4%