GLM-5.2: A Powerhouse for Aviation Operations Decision Support Evaluating Chain-of-Thought Reasoning in Complex Aviation Scenarios

GLM-5.2: A Powerhouse for Aviation Operations Decision Support — Frank Morales Aguilera, Sovereign Machine Laboratory (Montreal)

The paper evaluates the GLM-5.2 language model as a decision-support tool for airline Operations Control Centers. It uses a custom agentic system (the AOC_AGENTIC_GLM5DOT2_DEMO notebook) built on three components — a tool registry of 12 aviation-specific tools, an agent core, and a reflection layer — that synthesizes recommendations from structured operational data.

Method. Five scenarios were tested across passenger, cost, safety, and balanced-priorities conditions, with delays of 45–180 minutes, plus one replication. Each fed the model nine data dimensions (flight status, weather, crew rest, passenger impact, predictions, ATC, routes, regulatory compliance, cost). The model used a forced six-step reasoning process and was scored on seven metrics.

Results.

• 378–1,048 words of structured reasoning per scenario (average 708), with all six reasoning steps present in every case.
• Correctly identified FAR Part 117 and other regulatory violations in 100% of cases.
• Adapted recommendations to the stated priority (e.g., passenger → delay with amenities; safety → cancel immediately).
• Response times of 48–74 seconds.
• Positioned as outperforming GPT-4, Claude 3.5, and Gemini on structured output, regulatory knowledge, and aviation-domain expertise, and as approaching a senior operations manager with 20+ years of experience.

Conclusion. Recommends piloting GLM-5.2 in OCCs as a decision-support tool (with human oversight), and for training and compliance auditing. Stated limitations: simulated data only, no edge-case testing, no passenger-sentiment modelling, and the need for independent validation by aviation authorities. The implementation is available on GitHub.