#minicpm

0 stories this week · 2 this month · 4 in 6 months · mostly AI

Stories per day · 30 days

4 stories, newest first · one card per story

The publisher reports 93.68 on Real5-OmniDocBench and 1.2 pages per second on an RTX 5090.

10 min readminicpm ocr

Latent Compaction keeps 96.7% RULER accuracy at 32x KV-cache compression via test-time training.

LEAP retrieves short audio-visual windows for long recordings, beating Qwen3-Omni by up to 16.8%.

OpenBMB released MiniCPM5-2B, a dense 2B open-weights Transformer claiming 2B-class open-source SOTA for on-device deployment.

Stories tagged #minicpm