🤖 The Agents-A1 model has been introduced — a 35B MoE agent capable of competing with trillion-parameter models.
The key innovation lies in scaling the "agent horizon" (trajectory complexity) instead of simply increasing the number of parameters.
🌍 It demonstrates a new scaling path for agentic systems through the complication of logical chains and the distillation of expert knowledge.
👤 Highly efficient AI agents are becoming more accessible and faster, as they do not require massive computational power.
Source 1: https://internscience.github.io/Agents-A1/ Source 2: https://huggingface.co/InternScience/Agents-A1