LiquidAI: LFM2.5-24B-A2B

LFM2.5-24B-A2B, a mixture-of-experts reasoning model from Liquid AI, demonstrates strong reliability and a focus on agent-based tool calling. Created on October 13, 2026, this model, with 24B total and 2.45B active parameters, consistently provides usable responses, boasting a 90% success rate across benchmarks, indicating robust technical stability. However, its speed is a notable area for improvement, ranking in the 12th percentile for response times, suggesting it tends to be slower than many comparable models. Pricing is moderate, falling into the 33rd percentile. In terms of performance across categories, LFM2.5-24B-A2B shows a balanced, albeit average, accuracy. It achieved 61.0% in Instruction Following, 85.0% in Coding, 78.0% in Reasoning, and 89.0% in Mathematics. While its accuracy percentiles are generally in the mid-range (43rd-49th percentile), its performance in Mathematics and Coding is respectable. The model's core strength lies in its high reliability and its design for agent-tool interaction, which is supported by its "reasons before answering" approach. Its primary weakness is its extended response times across all evaluated benchmarks, which could impact real-time applications.