Introducing Atlas-2.1-Small
AeroAI launches its first official 3B-parameter model, with a 74.4% overall AvBench score, image input support, and local inference options.
Today, AeroAI is launching Atlas-2.1-Small, our first official 3-billion-parameter model. We chose Ministral 3 3B as a compact foundation for aviation question answering and conversational use: its 3B size offers a practical balance between capability and the memory and compute constraints of consumer hardware. That footprint also makes quantized local inference more practical than it would be with a much larger model, while giving us a foundation to adapt for aviation-focused behavior.
On AvBench 1.0, Atlas-2.1-Small scores 74.4% overall. Its category scores are Flight Planning Knowledge 72.7%; Flight Planning Computational Abilities 52.9%; Flight Plan Error Detection 46.2%; Aviation Formulas Knowledge 61.5%; Visual Chart Reading and Interpolation 83.3%; Non-Visual Chart Reading 100.0%; Aerospace Engineering 100.0%; Basic Aviation Math 100.0%; Complex Aviation Math 66.7%; and Airframe Systems and Formulas 90.9%. The results show strong performance on several knowledge, visual interpretation, and basic-math categories, alongside room to improve computational flight-planning tasks and error detection. Benchmark results describe performance on this evaluation set, not guaranteed reliability in real-world operations.
The model is available in BF16 and Q4_K_M quantization. Its intended uses include aviation knowledge questions, general aviation terminology and regulations, conversational assistance, and local deployment on supported edge devices. Atlas-2.1-Small also supports image input and more structured responses, including tables, lists, and formatted explanations.
Atlas-2.1-Small is part of the Atlas-2 model family. Atlas-2-Core and Atlas-2-Pro remain in development, while the earlier Atlas-2-Small is archived. The new release reflects our work to improve aviation knowledge, message understanding, explanation quality, and practical model deployment.
Atlas-2.1-Small is experimental and is not intended for commercial operational use. It may produce incorrect, outdated, or fabricated answers. Do not rely on it for flight planning, navigation, aircraft maintenance, or other safety-critical decisions. Verify aviation information against current FAA publications, NOTAMs, aircraft POH/AFMs, and qualified flight instructors. Fuel, weight-and-balance, and performance calculations must be independently checked.
Explore the model and its files on Hugging Face: https://huggingface.co/AeroAIAviation/Atlas-2.1-Small
