Cell · cited by the thesis

simple_qwen3.5-397b_basic_d1d4

Qwen3.5-397B on the simple floor, 5 episodes. Cited by the thesis in:

Outcome
Wins (boss defeated, alive)4/5 [0.38, 0.96]
Final HP / exact optimum, mean over wins15%
Died0
Final HP, mean112
Seconds per model call321.4
Parse error rate1.2%
Settings
FloorSimple floor
ModelQwen3.5-397B (qwen3.5-397b-a17b, thinking)
Planningreactive
Promptengine-default-floor-basic
Temperature0
Lookahead depth3
Mechanical overrideson (logged per turn)
Engine settings (environment, 1)
THINKING_MAX_TOKENS=32768
Episodesopen one to replay it turn by turn
#outcomefinal HPHP / optstepsmodel callsreplay
1won215%3617replay
2won215%3617replay
3won215%3617replay
4not won553–3616replay
5won215%3617replay
Re-run this cell

In a checkout of mt-ai-lab with an API key set (the result goes to a new file rerun_<time>_simple_qwen3.5-397b_basic_d1d4, never over the thesis result):

python scripts/run_catalog.py --codes B10

Result file: results/reliability/simple_qwen3.5-397b_basic_d1d4.json · updated 2026-09-26