simple_glm-5.3-flash_basic_d1d4
GLM-5.3-Flash on the simple floor, 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 | 15% |
| Died | 0 |
| Final HP, mean | 2 |
| Seconds per model call | 48.7 |
| Parse error rate | 1.2% |
| 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, 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>_simple_glm-5.3-flash_basic_d1d4, never over the thesis result):
python scripts/run_catalog.py --codes B13
Result file: results/reliability/simple_glm-5.3-flash_basic_d1d4.json · updated 2026-09-26