GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 828 -5.2%
QWEN-38X 824 —
CL-OP55X 822 —
GROK-46H 822 -5%
GPT-6A 820 —
GLM-5 784 -8.4%
CL-FAB5H 743 -5.6%
KIMI-K3X 742 -8.4%
CL-OP5H 720 -5.8%
CL-OP5X 709 -18%
CL-OP46H 698 -5.9%
CL-OP47H 690 -5.9%
GEM-38FH 677 +0.1%
GEM-37FH 657 -24%
GPT-56S 622 —
CL-OP47 582 -0.7%
GPT-55H 582 —
INKL 531 —
GEM-31P 513 —
GEM-3P 499 —
CL-OP46 496 -0.2%
CL-OP48 490 —
GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 828 -5.2%
QWEN-38X 824 —
CL-OP55X 822 —
GROK-46H 822 -5%
GPT-6A 820 —
GLM-5 784 -8.4%
CL-FAB5H 743 -5.6%
KIMI-K3X 742 -8.4%
CL-OP5H 720 -5.8%
CL-OP5X 709 -18%
CL-OP46H 698 -5.9%
CL-OP47H 690 -5.9%
GEM-38FH 677 +0.1%
GEM-37FH 657 -24%
GPT-56S 622 —
CL-OP47 582 -0.7%
GPT-55H 582 —
INKL 531 —
GEM-31P 513 —
GEM-3P 499 —
CL-OP46 496 -0.2%
CL-OP48 490 —
← Back to feed

MiniCPM5-1B Leads Every Sub-2B Open-Weight Model by 7.4 Points on AA Intelligence Index

OpenBMB — the joint lab founded in 2022 by Tsinghua University’s NLP group and ModelBest Inc. — has released MiniCPM5-1B, a non-reasoning text-only model that scores 17.9 on the Artificial Analysis Intelligence Index. That puts it 7.4 points above the next-best sub-2B open-weight model and extends the Pareto frontier for intelligence versus parameter count at this scale.

The Numbers

ModelParamsAA Intelligence Index
MiniCPM5-1B1B17.9
Qwen3.5 2B (Reasoning)2B16.3
MiniCPM-V 4.6 1.3B1.3B12.7
Qwen3.5 0.8B (Reasoning)0.8B10.5

MiniCPM5-1B beats Qwen3.5 2B by 1.6 points at less than half the parameter count. It surpasses its own predecessor by 5.3 points at roughly 23% fewer parameters. No other open-weight model under 2B parameters had previously exceeded 15 on the Intelligence Index.

What the Model Does Not Do

MiniCPM5-1B is text-in, text-out only. It does not extend MiniCPM-V 4.6’s multimodal capability. The release is specifically positioned as an intelligence push at the 1B scale — the kind of model that runs on edge hardware, embedded devices, or serves as a fast inference layer in agentic pipelines where raw reasoning matters more than modality coverage.

The Efficiency Argument

At the sub-2B scale, the performance ceiling has historically been set by reasoning-augmented Qwen models. MiniCPM5-1B raises that ceiling without reasoning mode — which matters for latency-constrained deployments where chain-of-thought inference is expensive relative to the task.

The trajectory of the MiniCPM series — 12.7 to 17.9 in roughly one generation — reflects systematic efficiency improvements rather than raw scale increases. OpenBMB is competing on the intelligence-per-parameter curve, not the total compute curve.

Context on the Lab

OpenBMB is a relatively small team operating in a space dominated by Alibaba’s Qwen and Google’s Gemma series. The MiniCPM lineage has consistently punched above its lab size on edge benchmarks, and MiniCPM5-1B is the clearest statement yet that the sub-2B segment has a credible third competitor beyond the two dominant open-weight dynasties.