MiniCPM5-2B-Jev prints 78.79% on JevBench, first among open models at or under 2B

OpenBMB posted MiniCPM5-2B-Jev on October 28, 2026, at 13:01 UTC. The project prints 78.79% on JevBench, first among open models at or under 2B, and 86.41% accuracy with ECE 0.0307 on decision-v7. Apache-2.0 on MiniCPM5-2B. This desk did not run the weights.

OpenBMB posted MiniCPM5-2B-Jev on October 28, 2026, at 13:01 UTC. The text names developer @ActbroDev and a System 1 decision model on OpenBMB’s MiniCPM5-2B that returns probabilities for predefined options without free-form generation. Reported results are 78.79% on JevBench, first among open models at or under 2B; 86.41% accuracy on the 10-source decision-v7 benchmark; and ECE 0.0307 on that suite. The post lists /v1/systemone, shared-context KV-cache reuse, and Apache 2.0 for code, LoRA adapters, and the calibration head. It shouts out yuting-ai. The share image on this page is that post’s still.

What the README prints

The GitHub README, opened October 31, prints JevBench overall 78.79% (table cell 78.8%): Easy 97.9%, Standard 94.4%, Hard 60.4% on 231 items. The same comparison table prints hosted Jev 1.13.0 at 86.6% and decider-2b v11 at 76.2%. On decision-v7 the project prints 86.41% accuracy, Brier 0.2143, and ECE 0.0307, against a printed TypeSafe Hosted Jev cell of 84.50% on that author table. The untuned MiniCPM5-2B base is 62.50% there. serve.py exposes /v1/systemone.

Weights

The Hugging Face card ytbai/MiniCPM5-2B-Jev, created September 28, 2026 and opened October 31, is Apache-2.0 on openbmb/MiniCPM5-2B. Ollama lists ActBro/minicpm5-2b-jev. No hosted POST on the pages this desk opened. This desk did not run the weights.

78.79% is JevBench overall. 86.41% and 0.0307 are decision-v7. Those stay on their own rows. OR-decision-4B’s 83.55% JevBench cell stays on /news/or-decision-4b/.

This site's reading

Editorial notes evaluating claims against primary sources, contextualizing findings alongside related implementations, and defining technical terms.

Verify

OpenBMB, @OpenBMB, October 28, 2026, 13:01 UTC, status 2115428701982908585. The text names developer @ActbroDev and MiniCPM5-2B-Jev on OpenBMB's MiniCPM5-2B, a System 1 decision model that returns probabilities for predefined options without free-form generation. Reported results: 78.79% on JevBench, ranking first among ≤2B open-weight System 1 / Jev models; 86.41% accuracy on the 10-source decision-v7 benchmark; ECE 0.0307 on that suite. The post lists /v1/systemone, shared-context KV-cache reuse, and Apache 2.0 for code, LoRA adapters, and the calibration head. It shouts out yuting-ai. The share image on this page is that post's still.

The GitHub README at yuting-ai/minicpm5-2b-jev, opened October 31, prints the same JevBench overall as 78.79% and 78.8% in the table (Easy 97.9%, Standard 94.4%, Hard 60.4% on 231 items). The decision-v7 table prints MiniCPM5-2B-Jev at 86.41% accuracy, Brier 0.2143, ECE 0.0307, against a printed TypeSafe Hosted Jev cell of 84.50% on that author table. The untuned MiniCPM5-2B base is 62.50% on that suite. The README serves /v1/systemone from serve.py.

The Hugging Face card ytbai/MiniCPM5-2B-Jev, created September 28, 2026, opened October 31. License apache-2.0. Base openbmb/MiniCPM5-2B. Ollama lists ActBro/minicpm5-2b-jev. No hosted POST on the pages this desk opened. This desk did not run the weights.

Compare

78.79% is the project's JevBench overall on 231 items, framed as first among open models at or under 2B. The same README table prints hosted Jev 1.13.0 at 86.6% on that comparison set. Do not merge 78.79 with Decision Index Full scores.

86.41% and ECE 0.0307 are decision-v7, the author's 10-source suite. The printed TypeSafe Hosted Jev cell on that table is 84.50%. That cell is not a JevBench composite and not TypeSafe's board rank.

OR-decision-4B's card JevBench 83.55% is a different pin and a 4B adapter. Jeva-4B's BANKING77 91.62 is a different suite. decider-2b v11 at 76.2% is the README's next ≤2B open row on that table.

Terms

MiniCPM5-2B-Jev: Community System One checkpoint on OpenBMB's MiniCPM5-2B, posted by OpenBMB on October 28, 2026. Project prints JevBench 78.79%, first among open models at or under 2B, and decision-v7 86.41% with ECE 0.0307. Apache-2.0. Weights ytbai/MiniCPM5-2B-Jev.
decision-v7: Author 10-source typed decision suite used in the MiniCPM5-2B-Jev README. The project prints 86.41% accuracy and ECE 0.0307 on that suite.