Mistral: Mistral Small 3.2 24B

Mistral-Small-3.2-24B-Instruct-2506, created on June 20, 2025, is an updated 24B parameter model from Mistral, optimized for instruction following, repetition reduction, and improved function calling. It supports image and text inputs with structured outputs and function/tool calling. The model performs among the fastest models, ranking in the top tier for speed (68th percentile), and consistently offers highly competitive pricing (83rd percentile). Demonstrating exceptional reliability, it achieved a 100% success rate across all benchmarks, indicating consistent and usable responses. In terms of performance, Mistral-Small-3.2-24B-Instruct-2506 exhibits strong capabilities in Ethics (100% accuracy), General Knowledge (98.5%), and Email Classification (99.0%), with the Ethics benchmark being particularly noteworthy for its perfect accuracy and efficiency. It also performs well in Hallucinations (96.0%) and Coding (86.0%). However, the model shows weaknesses in Instruction Following (51.0%) and Reasoning (60.0%), suggesting areas for improvement in complex multi-step directives and abstract problem-solving. Its Mathematics performance (85.0%) is moderate. Overall, its strengths lie in ethical adherence, general factual recall, and classification tasks, while complex instruction following and reasoning present challenges.