Google: Gemma 3n 4B

Gemma 3n 4B, released by Google on May 20, 2025, is designed for efficient execution on mobile and low-resource devices, supporting multimodal inputs and a wide linguistic range across 140+ languages. It demonstrates competitive response times, ranking in the 52nd percentile for speed, and consistently offers among the most competitive pricing, placing in the 93rd percentile. The model exhibits exceptional reliability with a 98% success rate, indicating minimal technical failures. In terms of performance across benchmarks, Gemma 3n 4B excels in Email Classification, achieving 99.0% accuracy and being noted as the most accurate model at its price point. It also performs well in Ethics with 98.0% accuracy. However, the model shows notable weaknesses in Instruction Following, with only 2.0% accuracy, and performs below average in General Knowledge (95.0%), Reasoning (58.0%), and Coding (71.0%). Its key strengths lie in its cost-effectiveness, high reliability, and strong performance in specific classification tasks, making it suitable for privacy-focused, on-device AI solutions where budget and consistent operation are critical.