MT3_qwen3.5-397b_basic_order_aware
Qwen3.5-397B on MT3, 5 episodes. Not cited by the thesis (development or earlier run).
| Wins (boss defeated, alive) | 5/5 [0.57, 1.00] |
| Final HP / exact optimum, mean over wins | 84% |
| Died | 0 |
| Final HP, mean | 377 |
| Seconds per model call | 19.9 |
| Parse error rate | 0.0% |
| Floor | MT3 |
| 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>_MT3_qwen3.5-397b_basic_order_aware, never over the thesis result):
python scripts/run_catalog.py --codes M3
Result file: results/reliability/MT3_qwen3.5-397b_basic_order_aware.json · updated 2026-09-25