Poolside: Laguna S 2.1 (free)

Poolside's Laguna S 2.1, a 118B total parameter model with 8B active parameters, demonstrates competitive response times, ranking in the 55th percentile across benchmarks. It typically offers cost-effective solutions, placing in the 70th percentile for pricing. Notably, the model exhibits strong reliability with a 95% success rate, indicating consistent and usable responses. In terms of performance across categories, Laguna S 2.1 shows particular strength in Classification, achieving 98.0% accuracy in Email Classification, making it the most accurate among models of comparable speed. It also performs well in Reasoning (90.0% accuracy) and General Knowledge (98.5% accuracy), indicating robust analytical and factual recall capabilities. Instruction Following is solid at 62.0% accuracy. However, a notable weakness appears in its Coding performance, where despite 85.0% accuracy, its duration is significantly longer (4004030ms), placing it in the 10th percentile for speed in this category. Hallucinations are relatively low at 90.0% accuracy, though this places it in the 36th percentile. Ethics performance is also a weaker area, with 83.0% accuracy ranking in the 17th percentile.