lang.ugrad.ai · MT engine arena · models first

Translation engines ranked · then languages they cover

Primary board is models that handle translation (qwen3 · m2m100 · mistral · …) with overall MT score and per-language H/M/L. Secondary tab is Alice R1 literary language slip/understanding (smoke test), not a substitute for the model list. Full matrix: /models.html.

Engine Arena · overall MT score

Score = operational MT fitness for zen fanout (routing), not COMET. Ready engines only scored.

loading mt-models.json…

Engine cards · language strengths

Semantic fidelity
Alice complete
Fluency
poet persona
Live usable
trunc smoke
Alice langs
R1 board
Default engine
qwen3:8b
offline

Alice language Arena% · slip / understanding smoke

Composite: 0.50·complete understanding + 0.30·live slip + 0.20·emit. This ranks languages under a fixed smoke, not which model is best.

loading arena-lang-board.json…

Full language table

RankLanguage Arena% Complete Live slip Emit Band Best engines

Language × engine (H / M / L)

Same matrix as models.html — which model supports which language.