Eagle-4 Released: Faster than DFlash2 & DSpark

Eagle-4 achieves 3.59x speedup on Qwen 3.8 27B, outperforming DFlash2 (2.80x) and DSpark (2.57x). Currently only supports Qwen 3.8 27B; community awaits integration into llama.cpp/vLLM.

• User builds 256GB VRAM rig with 8 Radeon Pro V620s2026-10-08
• Developer releases Ninfer 4080 for 16GB GPUs with 100k context2026-10-04
• Developer uses iPhone as second GPU to speed up MacBook LLM inference2026-10-02
• Four RTX 3060 Ti GPUs Hit 120 t/s Local Inference With 262K Context2026-09-27
• OpenBMB runs 35B MoE model locally on iPhone2026-10-21