Moondream: Moondream 3.1

Moondream 3.1, a 9B parameter (2B active) vision-language model, demonstrates exceptional speed and competitive pricing. It consistently ranks among the fastest models and offers highly competitive pricing across all evaluated benchmarks. The model's performance across specific benchmarks reveals areas for improvement. In Instruction Following, Moondream 3.1 achieved 21.0% accuracy, placing it in the 19th percentile, with moderate cost and duration. Its Coding performance was 3.0% accuracy (8th percentile), indicating a significant weakness in this domain, accompanied by higher cost and duration. The Reasoning benchmark yielded 0.0% accuracy, suggesting a substantial challenge for the model in complex problem-solving tasks. While its speed and cost-effectiveness are significant strengths, the model's accuracy in instruction following, coding, and reasoning tasks is notably low, indicating these areas as key weaknesses requiring further development.