Flight Planning Knowledge
Assesses end-to-end procedural flight planning knowledge, walking through route selection, operational constraints, waypoint sequences, and complete navlog creation.
Independent evaluation / Aviation AI
AvBench.
A benchmark for how frontier language models reason about aviation.
AvBench is an evaluation suite in development for flight planning, charts, aviation concepts, and complex math. It is designed to show capability, uncertainty, and failure modes—not just raw benchmark scores.
01 / What is AvBench?
More than a knowledge quiz.
AvBench tests whether a model can apply aviation concepts to new scenarios, identify when a critical input is missing, use evidence correctly, and distinguish a dependable answer from a confident but wrong one.
Each task can be evaluated without tools to measure unaided reasoning and with tools to measure how well a model retrieves, interprets, and uses evidence. AvBench is not an operational certification standard and should be understood as a research signal, not a flight-safety judgment.
02 / What we test
The aviation tasks that reveal reasoning.
Flight Planning Computational Abilities
Evaluates agentic planning of direct and cross-country flights using provided environmental factors like winds aloft and magnetic variation for precise grading.
Flight Plan Error Detection
Tests the model's ability to audit pre-filled navigation logs and flight plans to catch mathematical, operational, and route-consistency errors.
Aviation Formulas Knowledge
Examines formula recall and theoretical derivation across engineering principles, atmospheric conditions, and altitude conversions.
Visual Chart Reading & Interpolation
Measures multimodal visual understanding, reading graphical performance charts, sectionals, and interpolating points between published values.
Non-Visual Chart Reading
Tests data extraction and formulaic calculations derived from textual, tabular, and non-visual flight performance matrices.
Aerospace Engineering
Evaluates core aerospace principles, aerodynamics, propulsion mechanics, stability, and atmospheric physics without external retrieval.
Basic Aviation Math
Tests quick mental arithmetic, standard unit conversions, rule-of-thumb descent profiles, and basic speed-time-distance relationships.
Complex Aviation Math
Assesses multi-step mathematical operations including wind triangle vector resolution, weight and balance moments, and radius of turn calculations.
Airframe Systems & Formulas
Tests structural load factors, airframe stress limits, structural formulas, and physical system understanding without search tooling.
Interactive preview
A result surface built for comparison.
Hypothetical demonstration only — AvBench is still in development.
Hypothetical frontier LLM comparison
Fictional models, generated marks, and illustrative scores. These are not measured results.
Flight planning (w/Tools)
Direct, cross-country, and long-range planning with and without tools.
GPT-5.6 Sol (XHIGH)
78
Claude Sonnet 5 (MAX)
76
Gemini 3.8 Flash (HIGH)
71
Grok 4.6 (HIGH)
73
0255075100
03 / Access
Follow AvBench as it develops.
AvBench is currently in production. Contact AeroAI to ask about the benchmark, future methodology updates, or early access opportunities.
Contact AeroAI