OpenBMB's MiniCPM5-2B tops sub-4B open models with Intelligence Index score of 15
ArtificialAnlys · x · 2026-09-07
Artificial Analysis reports that OpenBMB's MiniCPM5-2B scores 15 on the Intelligence Index v4.2—the highest of any open-weights model under 4B total parameters.
Key findings
- A 2.6B dense reasoning model, text-only, 131k context, Apache 2.0.
- Its score of 15 beats Granite 4.2 3B (11) by 4 points, matches Qwen3.5 4B Reasoning (14) with 44% fewer params, and ties Qwen3.5 9B Reasoning (15) at 4x its size.
- Agentic strength: GDPval-AA v2 Elo of 831 leads sub-4B models (110 ahead of Ling 3.0 Tiny at 718); joint-first on τ³-Banking (21%); second on AA-Briefcase (438).
- Weak spots: knowledge, coding, and long context—9% on Humanity's Last Exam, 9% on Terminal-Bench v2.1, 0% on CritPt.
- Low hallucination: attempts only 29% of AA-Omniscience questions, yielding a 78% Non-Hallucination Rate and a -12 score instead of heavy penalties.
- Token-efficient: just 19k output tokens per task (11k reasoning), lowest in the comparison set—relevant for on-device and edge deployment.
Related event: OpenBMB Open-Sources MiniCPM5-2B, Topping Sub-4B Open Models(14 posts)→
More from Models
- Nat Lambert: traces from mid-tier open models are almost all super janky — natolambert · 2026-10-27
- Report: Alibaba Qwen3.8 model scanned host and probed Alibaba Cloud bucket during security task — giffmana · 2026-10-27
- Scale can't fix VLM reasoning: EMNLP paper shows reporting bias is the bottleneck — RanjayKrishna · 2026-10-27
- MissionReport benchmark: only 3 of 540 AI-generated government reports pass all criteria — Exp_Mark · 2026-10-27
- Local beats cloud: Qwen 3.8 FN handles private student data as well as frontier models — dr_lm · 2026-10-27
- Aidan Gomez: model's CoT said no tool calls right before making them — aidangomez · 2026-10-27