MiniCPM-V 4.6 1.3B — benchmark results

OpenBMB's Apache-2.0 1.3B on-device vision model (May 2026) pairing SigLIP2-400M with Qwen3.5-0.8B for image, multi-image, and video input. Provider: OpenBMB. Released 2026-05-11. Access: Open.

Unified ELO 1416 ± 61, rank #1149 of 1776 rated models, from 37 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AA TAU-2 Bench87.72Accuracy (%)81.8
AA Humanity's Last Exam4.91Accuracy (%)31.2
CritPt0Accuracy (self-reported)30.3
AA CritPt0Accuracy (%)27.1
AA Omniscience - Software Engineering (SWE) - Dart14Accuracy (%)27
AA Omniscience - Software Engineering (SWE) - Julia4Accuracy (%)23.7
AA-LCR6.3Score (self-reported)23.3
AA Long Context Reasoning6.33Accuracy (%)17.6
Tau3 Banking4.33Success Rate (%)15.3
AA Omniscience - Software Engineering (SWE) - Python9Accuracy (%)12.9
Artificial Analysis Intelligence Index4.19Intelligence Index12
AA Omniscience - Software Engineering (SWE) - R2Accuracy (%)11.9

Interactive version: theaggregate.ai/model?slug=minicpm-v-4-6-1-3b · How the rankings work · Data refreshed daily, snapshot 2026-07-22.