Is LearningUnit 2.0 (L1-30B-A5B) state of the art?Pending benchmarks.Announced Nov 6, 2026; no independent results yet.
Lunit and the Lunit Consortium's Korean-locale clinical foundation model, directly post-trained from Trillion Labs Gravity-30B-A5B-Base. The sparse MoE has 29.56B total and 5.34B active parameters, 131,072-token context, reasoning and retrieval/tool-call training. Apache-2.0 weights are publicly downloadable for self-hosting; no public hosted API is confirmed.
The details so far
- Lab
- Lunit
- Announced
- Nov 6, 2026
- API
- Not yet
- Open weights
- Published
- Parameters
- 29.56B total / 5.34B active
- Base
- Trillion Labs Gravity-30B-A5B-Base; GravityMoE architecture retained, with MLA, 52 layers, 64 routed experts (top-8) and one shared expert
- Capabilities
- Korean clinical locale and English; reasoning, multi-turn retrieval and tool-call trajectories with citation-grounded responses
- Context
- 131,072 tokens; BF16 checkpoint
- License
- Apache-2.0 verified in HF metadata, model card and repository LICENSE
- Serving
- Self-hosting instructions for Transformers and the Trillion Labs SGLang Gravity fork; no public hosted API stated. Published examples do not establish a hosted service
- Limitations
- Not a substitute for professional medical judgment; clinician review required. Knowledge may be outdated. Card requires repetition_penalty 1.05 to avoid repeated phrases on long answers; tool calling depends on schemas resembling the training retrieval harness, and citations need retrieved passages
- Related
- Separate from sibling L1-16B-A3B and Gravity-30B-A5B-Base; no sibling page added
- Date evidence
- Nov 6 official LearningUnit 2.0 press release. September repo creation/upload and a September title rename do not establish when 2.0 weights became public; historical first card called it Learning Unit 1 — 30B-A5B. No earlier public 2.0 availability verified
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: Lunit reports clinical and hospital-administration evaluations across nine medical institutions; these are attributed deployment/evaluation claims, not independent board results or a medical-use endorsement.
Sources
LABLunit announces LearningUnit 2.0lunit.io/ko/media-hub/루닛-의료-특화-파운데이션-모델-러닝유닛-2-0-공개/ · Nov 6, 2026REPOLearningUnit 2.0 L1-30B-A5B model card and weightshuggingface.co/learning-unit/L1-30B-A5BREPOL1-30B-A5B Apache-2.0 LICENSEhuggingface.co/learning-unit/L1-30B-A5B/blob/main/LICENSETracking continues until an independent board lists it.