#Eagle-4

Eagle-4 Released: Outpaces DFlash2 and DSpark in Inference Speed, 3.59× Speedup Measured on Qwen3.8-27B

Eagle-4 is officially released, achieving a 3.59× inference speedup on Qwen3.8-27B and comprehensively surpassing DFlash2 and DSpark. This article interprets the benchmark data, technical background, and llama.cpp/vLLM integration prospects.