MiniCPM5-1B: benchmark results
A dense 24-layer LlamaForCausalLM checkpoint with 1,080,632,832 total parameters, including 679,552,512 non-embedding parameters, and native 131,072-token context. The publisher explicitly supports Think and No Think modes through the same post-trained checkpoint; base and SFT-only releases remain separate. Provider: OpenBMB. Released 2026-05-21. Access: Open.
Aggregate Rating 1285 ± 29, rank #1485 of 1635 rated models, from 59 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AA-Omniscience Hallucination Rate | 0.9 | Hallucination Rate (%) | 100 |
| AA-Omniscience Index - Health | 0 | Omniscience Index | 86 |
| AA-Omniscience Index - Law | -0.1 | Omniscience Index | 81.1 |
| AA-Omniscience Index - Business | -1.1 | Omniscience Index | 80.8 |
| AA-Omniscience Index - Software Engineering (SWE) - Java | -1 | Omniscience Index | 79.7 |
| AA-Omniscience Index - Humanities & Social Sciences | -0.1 | Omniscience Index | 77 |
| AA Omniscience | -0.8 | Score | 75.7 |
| AA-Omniscience Index - Software Engineering (SWE) - Julia | 0 | Omniscience Index | 74.7 |
| AA-Omniscience Index - Software Engineering (SWE) - R | -2 | Omniscience Index | 73.6 |
| Tau2-Bench Telecom | 82.46 | Success Rate (%) | 73.3 |
| AA TAU-2 Bench | 82.46 | Accuracy (%) | 72.9 |
| AA-Omniscience Index - Science, Engineering & Mathematics | -2.5 | Omniscience Index | 72.8 |
Interactive version: theaggregate.ai/model?slug=minicpm5-1b · How It Works · Data refreshed daily, snapshot 2026-10-15.