complex_qwen3.5-397b_basic_k1k2
Qwen3.5-397B on the complex floor, 5 episodes. Not cited by the thesis (development or earlier run).
| Wins (boss defeated, alive) | 0/5 [0.00, 0.43] |
| Final HP / exact optimum, mean over wins | – |
| Died | 5 |
| Final HP, mean | -177 |
| Seconds per model call | 34.8 |
| Parse error rate | 1.3% |
| Floor | Complex floor |
| Model | Qwen3.5-397B (qwen3.5-397b-a17b, thinking) |
| Planning | reactive |
| Prompt | engine-default-floor-basic |
| Temperature | 0 |
| Lookahead depth | 3 |
| Mechanical overrides | on (logged per turn) |
Engine settings (environment, 1)
THINKING_MAX_TOKENS=32768
In a checkout of mt-ai-lab with an API key set (the result goes to a new file rerun_<time>_complex_qwen3.5-397b_basic_k1k2, never over the thesis result):
python scripts/run_catalog.py --codes CB4
Result file: results/reliability/complex_qwen3.5-397b_basic_k1k2.json · updated 2026-09-26