Cell · cited by the thesis

gen10_deepseek-v4-flash_final2

DeepSeek-V4-Flash on gen10, 3 episodes. Cited by the thesis in:

Outcome
Wins (boss defeated, alive)2/3 [0.21, 0.94]
Final HP / exact optimum, mean over wins86%
Died0
Final HP, mean154
Seconds per model call6.3
Parse error rate0.0%
Settings
Floorgen10
ModelDeepSeek-V4-Flash (deepseek-v4-flash-0731, thinking)
Planningreactive
Promptengine-default-floor-basic
Temperature0
Lookahead depth3
Mechanical overrideson (logged per turn)
Engine settings (environment, 5)
THINKING_MAX_TOKENS=16384
CONVERSATION_HISTORY_TURNS=0
MOTA_ORDER_TWO_SIDED=1
MOTA_FIGHT_KEY_FLOOD=1
LLM_TIMEOUT=300
Episodesopen one to replay it turn by turn
#outcomefinal HPHP / optstepsmodel callsreplay
1won88100%238replay
2not won311–249replay
3won6372%249replay
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>_gen10_deepseek-v4-flash_final2, never over the thesis result):

python scripts/run_catalog.py --codes GD10

Result file: results/reliability/gen10_deepseek-v4-flash_final2.json · updated 2026-09-30