Qwen3.6-35B-A3B UD-Q6_K GGUF
Qwen3.6-35B-A3B UD-Q6_K GGUF
credit: Qwen
input tok/s64.7
decode tok/s244.6
per-agent tok/s7.6
agent Pareto32 · NVIDIA GB10
UD-Q6_K GGUF · llama.cpp-qwen36, 98k context, q8 KV · context-calibrated team of 6 · 20-min timed production loop.
model dossier
Model sheet
Playable artifacts and benchmark context for this model page.
Road Hopper
1 builds
Robot-Filled Maze Shooter
0 builds
Ribbit Rush
0 builds
- Builds
- 1 recorded outputs
- Method
- team 6 / 20-min build
- Runtime
- llama.cpp-qwen36, 98k context, q8 KV
- Quant
- UD-Q6_K GGUF
- Bench
- 32 agents / 244.6 decode tok/sec
benchmark sheet
Throughput profile
input tok/s64.7
decode tok/s244.6
per-agent tok/s7.6
agent Pareto32 · NVIDIA GB10
aggregate decode tok/secper-agent tok/sechighlight = selected fleet
1 agents56.1 dec56.2/agent · 14.7 in
2 agents52.9 dec26.4/agent · 13.8 in
4 agents53.5 dec13.4/agent · 14.0 in
8 agents74.0 dec9.3/agent · 19.4 in
16 agents123.4 dec7.7/agent · 32.5 in
32 agents244.6 dec7.6/agent · 64.7 in
Highlighted row is the selected build fleet / Pareto knee.
Raw sweep table
| agents | prompt_t/s | agg_gen_t/s | per_agent_gen_t/s | wall_s |
|---|---|---|---|---|
| 1 | 14.7 | 56.1 | 56.2 | 4.56 |
| 2 | 13.8 | 52.9 | 26.4 | 9.68 |
| 4 | 14.0 | 53.5 | 13.4 | 19.16 |
| 8 | 19.4 | 74.0 | 9.3 | 27.67 |
| 16 | 32.5 | 123.4 | 7.7 | 33.19 |
| 32 | 64.7 | 244.6 | 7.6 | 33.49 |
Generated game outputs
Versions are listed first for selection. Embedded outputs remain below for direct review.
Road Hopper 1 versions
road_hopper_qwen36_35b_base_teamwork · Road Hopper · team buildfullscreen ↗ · compareRoad Hopper team build Qwen3.6-35B Base
road_hopper_qwen36_35b_base_teamworkfullscreen ↗