Meet AntAngelMed: A 103B-Parameter Open-Source Medical Language Model Built on a 1/32 Activation-Ratio MoE Architecture

Short summary

MedAIBase released AntAngelMed, an open-source 103B-parameter medical language model using sparse MoE to activate only 6.1B parameters at inference, achieving 200+ tokens/sec on H20 hardware. It ranks first among open-source models on HealthBench, MedAIBench, and MedBench. Built through continual pre-training, supervised fine-tuning, and GRPO-based reinforcement learning.

• •103B-parameter medical LLM using sparse MoE architecture (6.1B active parameters)
• •Matches 40B dense model performance at 200+ tokens/sec on H20
• •Ranks #1 on HealthBench, MedAIBench, and MedBench leaderboards

Generated with AI, which can make mistakes.