MoonshotAI: Kimi K3

MoonshotAI's Kimi K3, an ultra-large-scale, open-weight multimodal reasoning model, demonstrates exceptional reliability with a 100% success rate across all benchmarks, consistently providing usable responses. However, it tends to have longer response times, ranking in the 16th percentile for speed, and is positioned at premium pricing levels, ranking in the 6th percentile for cost. Despite its higher cost and slower speed, Kimi K3 exhibits strong performance in several key areas. It achieved perfect accuracy in both General Knowledge and Ethics benchmarks, notably being the most accurate model at its price point and among models of comparable speed in these categories. Its instruction following capabilities are robust, scoring 82.0% accuracy, placing it in the 86th percentile. The model also shows strong reasoning abilities with 98.0% accuracy (84th percentile) and effectively acknowledges uncertainty, achieving 98.0% accuracy in the Hallucinations benchmark. Its primary weakness appears to be in Email Classification, where its 97.0% accuracy places it in the 35th percentile, indicating room for improvement compared to other models. Overall, Kimi K3 is a highly reliable model with significant strengths in complex reasoning, knowledge, and ethical considerations, albeit at a premium price and with slower processing.