Minicpm5 1b The Leading 1b Open Weights Model

OpenBMB, a China-based lab founded in 2022 by Tsinghua University's NLP Lab and ModelBest Inc., has released MiniCPM5-1B, a dense 1B-parameter model that takes text in and puts text out, with open weights under an Apache 2.0 license. Artificial Analysis scores the non-reasoning model 17.9 on its Intelligence Index, which it reports is 7.4 points ahead of any other open-weights model at 1B parameters or below; the next is Qwen3.5 0.8B (Reasoning) at 10.5 (artificialanalysis.ai). It also beats Alibaba's Qwen3.5 2B (Reasoning, 16.3) by 1.6 points with less than half the parameters, and no other open-weights model under 2B parameters has exceeded 15.

The jump is measured against OpenBMB's own MiniCPM-V 4.6 1.3B, released two weeks earlier and placed at 12.7 here. MiniCPM5-1B adds 5.3 points with about 23% fewer parameters, but drops that model's image and video input. It is also more talkative: it used 12.6M output tokens to run the Index, about 2.3 times the predecessor's 5.4M, though still about 31 times fewer than Qwen3.5 2B (Reasoning, 389M) and about 8 times fewer than Qwen3.5 2B (Non-reasoning, 100M).

AA-Omniscience, Artificial Analysis's knowledge and hallucination test, penalizes wrong answers and credits a model positively for declining instead. Sub-2B models usually attempt most questions and hallucinate often, which puts peers such as Qwen3.5 0.8B (Non-reasoning) at -89, MiniCPM-V 4.6 1.3B at -85 and Exaone 4.0 1.2B (Non-reasoning) at -83. MiniCPM5-1B declines the vast majority of questions and scores -1, the highest in its size class. Artificial Analysis attributes that score to abstaining rather than hallucinating, so it reflects caution rather than greater recall. No inference providers were confirmed at release.