Tag
Liquid AI has released Liquid Nanos, a family of small foundation models (350M–2.6B parameters) that deliver frontier-grade performance on specialized tasks while running on everyday devices.
The paper introduces Hierarchical Self-Improvement (HSI), a framework that enhances frozen LLM agents by evolving task-specific harnesses through hierarchical self-modification, achieving substantial gains on moderate tasks while being limited by feedback quality and backbone capabilities.
Model routing is a hot trend to reduce inference costs, but the best routing is deeply task-specific. Teams like Harvey and Factory achieve significant cost savings by focusing on single workflows rather than generic routers.
The author trained a Qwen3.6-35B-A3B model using reinforcement learning to then RL-train small task-specific Qwen models, and has released everything fully open source.
A 6-person team built task-specific AI models that are 4-8x faster than OpenAI or Anthropic models, with 500K downloads on HuggingFace.