NeoMME gets within 0.002 nDCG of ColQwen2.5 at 14x fewer parameters.

H Company released it September 3, a multimodal-native encoder at 260M and 800M running text tokens and raw image patches through one bidirectional Transformer with no separate vision tower or text decoder, using a from-scratch 131k BPE vocabulary, 16,384-token context, 32x32 dynamic-resolution patches, GQA and 2D RoPE. On ViDoRe v3 the 260M scores 0.523 nDCG@10 and the 800M reaches 0.556, with throughput at 51 pages/second at 2048x2048 on an L40S, about double ColModernVBERT. All checkpoints Apache 2.0, so it's a cheap swap for document RAG (Hugging Face).