Is Numo-1-E2B state of the art?Pending benchmarks.Announced Oct 21, 2026; no independent results yet.
Mycel's small open model for decisions about scientific papers. Given a title and abstract, it answers a fixed question, such as study design, research field or relevance to a seed paper, with a probability for every answer, in one pass on a consumer GPU. Fine-tuned from Gemma 4 E2B; Apache 2.0; English only; a research release.
The details so far
- Lab
- Mycel
- Announced
- Oct 21, 2026
- API
- Not yet
- Open weights
- Published
- Parameters
- Gemma 4 E2B base (about 2B effective); 4.6B parameters in the safetensors files
- Base model
- google/gemma-4-E2B-it (fine-tune)
- Tasks
- Study approach, research area, biomedical relevance, arXiv field, hypothesis support and annotator consensus
- Limits
- Inputs capped at 1,024 tokens; English only; weak on general benchmarks
- Speed
- About 124 ms per decision on one RTX 3060, offline (lab measurement)
- License
- Apache 2.0 weights and code; DOI 10.57967/hf/11188
Tags marked unsure are not stated on the lab's own pages.
What would let us call it
Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.
- Publicly announced with a source
- Usable by the public (API or weights)
- Listed on an independent board we track
- Enough results for a rank in a category
- Holds across two snapshots (crowns only)
What the lab claims
Self-reported. Shown for context; not used in any verdict.
Lab claim: 94.3% study-design accuracy on 600 recent PubMed papers labelled by NLM indexers, within 2 points of the best hosted decision models in Mycel's comparison, with the lowest calibration error (1.6%) of the models tested.
Lab claim: 98% accuracy on the 80% of papers it is most confident about.
Sources
REPONumo-1-E2B on Hugging Facehuggingface.co/mycel-ai/Numo-1-E2B · Oct 21, 2026LABAlphin Tom (Mycel CEO) announcement on Xx.com/alphinctom/status/2112945003852914968 · Oct 21, 2026Tracking continues until an independent board lists it.