Compact public methodology
How Loquela labels model evidence
The public catalog is designed to be useful without turning estimates or upstream marketing into Loquela benchmark results. Detailed internal evaluation plans, product strategy, and unreleased research remain private.
Measured
Captured by Loquela with a recorded device, build, model package, sample set, and timing definition.
Upstream
Published by the model, runtime, Android, or package maintainer and linked to that primary source.
Estimated
A planning value derived from a stated heuristic, not presented as observed performance.
Not yet measured
A useful field for which Loquela does not yet have defensible comparable evidence.
Minimum comparison record
A published device result should identify the Loquela version, phone model, Android version, processor and memory where available, exact model ID, audio sample set, test repetitions, and whether timing covers only inference or the complete stop-to-text experience.