TheDrummer: Anubis Pro 105B V1

TheDrummer's Anubis Pro 105B V1 demonstrates competitive response times, ranking in the 46th percentile across six benchmarks, and offers competitive pricing, placing in the 42nd percentile. This model, an expanded variant of Meta’s Llama 3.3 70B, is designed for advanced narrative, roleplay, and instructional tasks. Performance across benchmarks reveals a mixed profile. Anubis Pro 105B V1 excels significantly in Email Classification, achieving perfect 100% accuracy, making it the most accurate model at its price point and among models of similar speed. It also shows moderate capability in Instruction Following, scoring 53% accuracy. However, the model exhibits notable weaknesses in other areas. Its General Knowledge accuracy is low at 35%, and it struggles considerably with Reasoning (8% accuracy) and Coding (19% accuracy). Ethics performance is also relatively low at 68%. While its larger parameter count aims for deeper contextual understanding, this does not consistently translate to strong performance across all complex cognitive tasks. The model's strengths lie in specific classification and instruction adherence, while general knowledge, complex reasoning, and coding remain areas for improvement.