Minicpm5-2b, out september 7, matches gemma 4 12b on benchmarks with 2 billion parameters, and…
minicpm5-2b, out september 7, matches gemma 4 12b on benchmarks with 2 billion parameters, and openbmb shipped the full stack with it — datasets, training recipes, reinforcement learning infra. the race is no longer bigger models, it is more intelligence per parameter.
the weights were the least interesting thing in that release.
Context
The MiniCPM5-2B model card says it is a dense 2B Transformer and a 2B-class open-source state of the art within its comparison set, with vendor table averages of 53.9 for MiniCPM5-2B, 51.1 for Qwen3.5-4B, 42.7 for granite-4.2-3B, 31.2 for Gemma-4-E4B-it and 24.6 for Gemma-4-E2B-it. Artificial Analysis, 7 September 2026, says it is a 2.6B-parameter dense reasoning model under Apache 2.0 with a score of 15 on its Intelligence Index v4.2, the highest of any open-weights model under 4B total parameters, and that on Humanity's Last Exam it scores 9%, behind Gemma 4 12B (Reasoning) at 16%.
The vendor table averages are vendor-reported over the vendor's chosen tasks. Gemma 4 12B does not appear in the vendor table, which uses Gemma-4-E2B-it and E4B, and Artificial Analysis shows MiniCPM5-2B behind Gemma 4 12B on HLE, so matches Gemma 4 12B on benchmarks is not supported by the rows read. The parameter count is 2.6B per Artificial Analysis versus 2B-class on the card. The full stack of datasets, training recipes and RL infrastructure was not found in the lines read. The model card's release date was not inspected. The two benchmark sets are not merged.
Watch next
- The model card release date and the stack contents.
Sources
- MiniCPM5-2B model card (Hugging Face)huggingface.co
- OpenBMB releases MiniCPM5-2B (Artificial Analysis, 7 Sep 2026)artificialanalysis.ai
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 18 September 2026 at 16:51 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →