Tencent: Hunyuan A13B Instruct

Tencent's Hunyuan A13B Instruct model, a 13B active parameter Mixture-of-Experts (MoE) model with 80B total parameters, demonstrates strong overall performance, particularly in speed and cost efficiency. It consistently ranks among the fastest models and offers highly competitive pricing across various benchmarks. The model also exhibits strong reliability, achieving a 95% success rate. In terms of benchmark performance, Hunyuan A13B shows notable strengths in Reasoning, where it achieved 96.0% accuracy and was identified as the most accurate model at its price point. It also performed well in General Knowledge (97.5% accuracy) and Coding (90.0% accuracy). However, a significant weakness is observed in Instruction Following, where it registered 0.0% accuracy across two separate tests, indicating a critical area for improvement. Mathematics performance was moderate at 59.0% accuracy, and Email Classification was 91.0% accurate but ranked in the 19th percentile, suggesting it's less competitive in this specific task. The model's support for Chain-of-Thought reasoning, GQA, and quantization (FP8, GPTQ) contributes to its efficiency and competitive edge.