Goliath 120B

Goliath 120B, an alpindale model created by merging two fine-tuned Llama 70B models (Xwin and Euryale), consistently ranks among the fastest models and offers highly competitive pricing across various benchmarks. Its speed and cost-efficiency are significant advantages. In terms of performance, Goliath 120B demonstrates a notable strength in classification tasks, achieving 95.0% accuracy in Email Classification, placing it in the 34th percentile for this category. This suggests strong capabilities in discerning context and categorizing information. However, the model exhibits significant weaknesses in other areas. Its performance in Instruction Following is particularly concerning, with 0.0% accuracy, indicating a fundamental limitation in processing and executing complex multi-step instructions. Similarly, its accuracy in Coding (22.0%), Reasoning (39.0%), and Ethics (44.0%) benchmarks is relatively low, placing it in the lower percentiles for these categories. While its cost performance is generally strong across all benchmarks, the low accuracy in critical areas like coding and reasoning suggests that while it's cost-effective, its utility for tasks requiring high precision in these domains may be limited.