AionLabs: Aion-1.0-Mini

AionLabs: Aion-1.0-Mini consistently performs among the fastest models and offers highly competitive pricing, ranking in the Infinityth percentile across six benchmarks for both speed and cost. This 32B parameter model, a distilled version of DeepSeek-R1, is designed for strong performance in reasoning domains. In terms of specific benchmark performance, Aion-1.0-Mini demonstrates notable strengths in Email Classification, achieving 99.0% accuracy (81st percentile), indicating excellent contextual understanding for categorization tasks. Its Reasoning capabilities are also strong, with 82.0% accuracy (82nd percentile), aligning with its design for logic and problem-solving. However, the model exhibits significant weaknesses in other areas. Its Instruction Following accuracy is 0.0%, suggesting a critical limitation in processing and executing complex directives. Performance in Ethics (41.0% accuracy) and Coding (36.0% accuracy) is also low, placing it in the lower percentiles for these categories. General Knowledge is moderate at 49.3% accuracy. While its speed and cost are exceptional, the variability in accuracy across different domains highlights areas for potential improvement, particularly in instruction adherence and ethical reasoning.