LiquidAI: LFM3-1.7B

LiquidAI’s LFM3-1.7B shows moderate speed performance, ranking in the 28th percentile, while offering competitive pricing situated at the 56th percentile across evaluated benchmarks. The model also demonstrates strong reliability, achieving an 86% technical success rate with very few operational failures. Across benchmark categories, LFM3-1.7B delivers solid performance in specialized technical domains despite its compact 1.7B parameter footprint. Its primary strengths lie in mathematics and coding, achieving 88.0% accuracy on complex math evaluations and 82.0% accuracy across multiple programming languages. It also maintains competent deductive capabilities, posting a 70.0% accuracy in multi-step reasoning tasks. Conversely, the model’s notable weakness is in complex instruction following, where its accuracy dropped to 54.0%, indicating that it may struggle with intricate, multi-layered conditional directives or strict formatting constraints. While its accuracy percentiles relative to the broader model landscape sit in the mid-to-high 30s, LFM3-1.7B couples its technical problem-solving capabilities with a massive 256K-token context window and internal reasoning mechanisms. This makes it a reliable, cost-effective choice for developers seeking an efficient, lightweight model for math and code-oriented agentic workflows.