Liquid: LFM 40B MoE

Liquid's LFM 40B MoE model, launched on September 29, 2024, is a 40.3B Mixture of Experts model built on dynamic systems, designed for general-purpose sequential data modeling. This model consistently ranks among the fastest available, demonstrating exceptional speed performance across five benchmarks. It also offers highly competitive pricing, placing it among the most cost-effective options in the market. In terms of specific benchmark performance, the LFM 40B MoE shows a mixed profile. It exhibits strong performance in Email Classification and Ethics, achieving 96.0% and 95.5% accuracy respectively, indicating a solid grasp of contextual understanding and ethical reasoning. Its Instruction Following capability is moderate at 44.4% accuracy. However, the model struggles significantly with General Knowledge, scoring only 17.2% accuracy, and completely fails in the Coding benchmark with 0.0% accuracy. This suggests a notable weakness in factual recall and programming comprehension. Its cost efficiency is generally good across benchmarks, and its duration for tasks is varied, being particularly fast for Ethics and Email Classification.