Cell · cited by the thesis

simple_qwen3.5-397b_basic_k1k2_hist0

Qwen3.5-397B on the simple floor, 10 episodes. Cited by the thesis in:

Outcome
Wins (boss defeated, alive)9/10 [0.60, 0.98]
Final HP / exact optimum, mean over wins22%
Died0
Final HP, mean58
Seconds per model call56.5
Parse error rate0.0%
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, 2)
THINKING_MAX_TOKENS=32768
CONVERSATION_HISTORY_TURNS=0
Episodesopen one to replay it turn by turn
#outcomefinal HPHP / optstepsmodel callsreplay
1won215%3617replay
2won215%3617replay
3won1077%3616replay
4won215%3617replay
5won215%3617replay
6won215%3617replay
7won215%3617replay
8won215%3617replay
9won215%3617replay
10not won550–3617replay
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_k1k2_hist0, never over the thesis result):

python scripts/run_catalog.py --codes H0 H0b

Result file: results/reliability/simple_qwen3.5-397b_basic_k1k2_hist0.json · updated 2026-09-26