Atlas

Models

← All models

MiniCPM5-1B

OpenBMB · 2026-05-25 · 1.1B parameters

MiniCPM5-1B is OpenBMB's dense 1B on-device language model and the first checkpoint in the MiniCPM5 series. Its training spans base training, mid-training, and SFT, RL, and on-policy distillation from specialized RL teachers; the released checkpoint supports switchable Think and No Think modes, a 131,072-token context window, and XML-style tool calling for local assistants and resource-constrained deployments.

Benchmark scores

BenchmarkScore
AA-LCR4.7
AA-Omniscience Index-0.8
Artificial Analysis Intelligence Index12.0
Artificial Analysis Omniscience Accuracy1.7
Artificial Analysis Omniscience Hallucination Rate0.9
Artificial Analysis Openness Index83.3
Capability116.2
CritPt0.0
GPQA Diamond27.8
Humanity's Last Exam (Text-Only)6.5
IFBench (Artificial Analysis)49.3
SciCode (Artificial Analysis)4.4
Terminal-Bench Hard0.0
τ²-bench (Telecom)82.5
Loading Atlas data…