# FrontiersMind releases 27-billion-parameter Lumma-Fev decision model

> The model chooses among supplied options rather than generating prose; its developer reported 90.7% accuracy on 4,454 test examples.

- Canonical URL: https://newsverum.com/en/news/frontiersmind-lumma-fev-27b-open-decision-model
- Publisher: Newsverum
- Section: Technology
- Published: 2026-10-15T20:19:53+00:00
- Updated: 2026-10-15T20:19:53+00:00
- Event / data period: 2026-10-14
- Other editions: [Türkçe](https://newsverum.com/tr/news/frontiersmind-lumma-fev-27b-open-decision-model.md), [العربية](https://newsverum.com/ar/news/frontiersmind-lumma-fev-27b-open-decision-model.md), [Français](https://newsverum.com/fr/news/frontiersmind-lumma-fev-27b-open-decision-model.md), [Español](https://newsverum.com/es/news/frontiersmind-lumma-fev-27b-open-decision-model.md), [Deutsch](https://newsverum.com/de/news/frontiersmind-lumma-fev-27b-open-decision-model.md), [Português](https://newsverum.com/pt/news/frontiersmind-lumma-fev-27b-open-decision-model.md)

FrontiersMind released Lumma-Fev-27B on October 14. The decision model accepts a state and a set of choices, then returns a selected label and probabilities rather than free-form text. The developer says it trained all weights on about 100,000 examples and supports English, Hindi, Tamil and Kannada.

In the developer’s test on 4,454 held-out decisions, Lumma-Fev-27B scored 90.7% accuracy versus 83.9% for TypeSafe Jev, with reported calibration errors of 0.009 and 0.045. The developer says Jev still leads on broader reasoning tests such as BBH and WinoGrande. These are developer-run results, not an independent evaluation.

## Data

### FrontiersMind releases 27-billion-parameter Lumma-Fev decision model

Methodology: This structured record summarizes the cited announcement; it is not an independent performance evaluation.

| date | label | description | status |
|---|---|---|---|
| 2026-10-14 | FrontiersMind releases 27-billion-parameter Lumma-Fev decision model | The model chooses among supplied options rather than generating prose; its developer reported 90.7% accuracy on 4,454 test examples. | occurred |

Sources: <https://huggingface.co/blog/FrontiersMind/lumma-fev-27b> · Download data: [JSON](https://newsverum.com/api/v1/articles/frontiersmind-lumma-fev-27b-open-decision-model/data/frontiersmind-lumma-fev-27b-open-decision-model-release?lang=en&format=json) · [CSV](https://newsverum.com/api/v1/articles/frontiersmind-lumma-fev-27b-open-decision-model/data/frontiersmind-lumma-fev-27b-open-decision-model-release?lang=en&format=csv)

## Sources

- [FrontiersMind on Hugging Face Blog](<https://huggingface.co/blog/FrontiersMind/lumma-fev-27b>) (Source checked: 2026-10-15; Published: 2026-10-14)

---

Cite the canonical URL above. [Editorial standards and corrections](https://newsverum.com/en/editorial)