bench.karti.ai
← board

Qwen3.6-35B-A3B (NVFP4)

qwen3.6-35b-a3b-nvfp4@spark-1

NVIDIA GB10 Grace Blackwell (DGX Spark)35B·3B activeNVFP4ctx 65,536measured 2026-07-28 06:24Z
Capped run. Reference tasks were limited to 100 samples. Not comparable to a full-dataset score.

Quality

ifevalreference
0.936
100 samples2215sprompt_strict_acc 0.930prompt_strict_stderr 0.026prompt_loose_acc 0.920prompt_loose_stderr 0.027
per-sample outcomes · 93/100 passed

Serving performance

921852773691832TOK/S AGGREGATE ↑ / CONCURRENCY →
conctok/s all/streamttft p50ttft p95tpot
127.627.6321ms321ms35.2ms
8174.521.91929ms2069ms38.4ms
32320.910.46087ms9160ms78.3ms
serving configuration

Results are only comparable between targets with identical flags. These are recorded from the live server, not from memory.

enforce_eager
true
gpu_memory_utilization
0.55
kv_cache_dtype
fp8
max_num_batched_tokens
4096
moe_backend
flashinfer_b12x
reasoning_parser
qwen3
tool_call_parser
qwen3_coder
enable_auto_tool_choice
true
notes

Live behind the `brain` and `115-model-local-moe` aliases. This is the model the bench exists to interrogate: it was deployed without measurement. FLAGS OF CONCERN: --enforce-eager disables CUDA graphs (costs throughput) and --gpu-memory-utilization 0.55 leaves ~45% of unified memory unused, capping KV cache and therefore concurrency. Both are worth an A/B once the baseline is recorded.

download score card ↓PNG, sized for sharing. Same image the link preview uses.