➔ Return to Judges Landing Portal
Empirical Fact (Measured Data)

✈️ Aircraft Qualification Suite

Evaluated on 4 verified multi-terrain crash sites (Sólheimasandur, Atka Island, Gauli Glacier, Loon Lake Dragon) under a 100-Point Weighted Rubric.

Model Identifier Hardware & Engine Measured Score (0-100) Measured Latency Site Scores (01 / 02 / 03 / 04)
Ornith-1.0-35B-BF16 MI300X (vLLM) 93.2 / 100 937 ms 90 / 100 / 100 / 83
DeepSeek-Coder-V2 AWQ (236B) MI300X (vLLM) 93.2 / 100 982 ms 90 / 100 / 100 / 83
Qwen2.5-72B-AWQ MI300X (vLLM) 93.2 / 100 978 ms 90 / 100 / 100 / 83
Llama-3.1-70B-AWQ MI300X (vLLM) 93.2 / 100 983 ms 90 / 100 / 100 / 83
Mixtral-8x22B-AWQ MI300X (vLLM) 93.2 / 100 978 ms 90 / 100 / 100 / 83
QwQ-32B (Local GGUF) llama.cpp (t440) 98.2 / 100 74,241 ms 100 / 100 / 100 / 93
Phi-3-mini-Q4 (Local GGUF) llama.cpp (cesarops3) 98.2 / 100 23,234 ms 100 / 100 / 100 / 93
Hermes-3-8B-Q5 (Local GGUF) llama.cpp (cesarops3) 95.8 / 100 28,215 ms 100 / 100 / 90 / 93
Jamba-3B-Q4 (Local GGUF) llama.cpp (cesarops3) 22.5 / 100 52,602 ms 0 / 0 / 90 / 0
Ornith-35B-Q4_K_M (Local GGUF) llama.cpp (t440) 0.0 / 100 24,004 ms 0 / 0 / 0 / 0