The premium
AI models read, bill and limit text in tokens. For the same meaning, African languages need more tokens than English. That premium is paid on every query, in cost, in how much text a model can read at once, and in speed.
The stack
Nzinga is three parts: a tokenizer that cuts the premium, the Nzinga Network where members record their own languages, and the phone app they record with. Members choose who can use every recording.
nzinga:
tokenizer: gemma3-afx32k
vocabulary: 294,145 tokens
network: invite-only
app: Expo and React Native
Your language. Your voice. Your terms.
Brand line
,
Nzinga
The result
Nzinga needs fewer tokens than all 12 other tokenizers tested on all 12 Tier 1 languages: 22% fewer than the best. Whether models trained with it do better is not measured yet.
