Laguna S 2.1 122B vs Ornith 35B vs Qwen 3.6 35B

KAT-Coder V2.5 vs Ornith 35B vs Qwen 3.6 35B vs Laguna S 2.1 122B — each model got one shot to generate a CS-style FPS game. Only 1 attempt per model.

1st
Ornith 1.0 35B-A3B UD-Q8_K_XL

huggingface.co/unsloth/Ornith-1.0-35B-GGUF

Cleanest output. Working FPS prototype with full HUD, crosshair, movement, and bots.

2nd
KAT-Coder V2.5 Dev Q8_0

huggingface.co/bartowski/Kwaipilot_KAT-Coder-V2.5-Dev-GGUF

Strong coding model. Solid FPS output with good structure.

3rd
Qwen 3.6 35B-A3B UD-Q8_K_XL

huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF

Solid attempt with good visuals. Required a fix to become fully playable.

4th
Laguna S 2.1 UD-Q4_K_XL

huggingface.co/unsloth/Laguna-S-2.1-GGUF

Unplayable. Minimal output that didn't produce a working game.

Prompt
Loading prompt...
Run Parameters (llama.cpp router)
Parameter Laguna S 2.1 UD-Q4_K_XL Ornith 35B UD-Q8_K_XL KAT-Coder V2.5 Dev Q8_0 Qwen 3.6 35B-A3B UD-Q8_K_XL
ctx-size 262144 262144 262144 262144
n-gpu-layers all all all all
n-cpu-moe 46 40 29 40
kv-offload true true true true
cache-type-k/v q8_0 q8_0 q8_0 q8_0
temp 0.7 0.6 0.6 0.6
top-p 1.0 0.95 0.95 0.95
top-k 20 20 20 20
min-p 0.0 0.0 0.0 0.0
repeat-penalty 1.0
reasoning on on on on
spec-type draft-mtp
spec-draft-n-max 2
dry-multiplier / base 0.6 / 1.75