simple_glm-5.3-flash_final2_la3
GLM-5.3-Flash on the simple floor, 2 episodes. Not cited by the thesis (development or earlier run).
| Wins (boss defeated, alive) | 0/2 [0.00, 0.66] |
| Final HP / exact optimum, mean over wins | – |
| Died | 0 |
| Final HP, mean | – |
| Seconds per model call | – |
| Parse error rate | – |
| Floor | Simple floor |
| Model | GLM-5.3-Flash (glm-5.3-flash, thinking) |
| Planning | reactive |
| Prompt | engine-default-floor-basic |
| Temperature | 0 |
| Lookahead depth | 3 |
| Mechanical overrides | on (logged per turn) |
Engine settings (environment, 5)
THINKING_MAX_TOKENS=32768 CONVERSATION_HISTORY_TURNS=0 MOTA_ORDER_TWO_SIDED=1 MOTA_FIGHT_KEY_FLOOD=1 LLM_TIMEOUT=300
| # | outcome | final HP | HP / opt | steps | model calls | replay |
|---|---|---|---|---|---|---|
| 1 | not won | – | – | – | – | log not stored |
| 2 | not won | – | – | – | – | log not stored |
In a checkout of mt-ai-lab with an API key set (the result goes to a new file rerun_<time>_simple_glm-5.3-flash_final2_la3, never over the thesis result):
python scripts/run_catalog.py --codes GLS
Result file: results/reliability/_error_only/simple_glm-5.3-flash_final2_la3.json · updated 2026-10-01