Releases
What we measured,release by release.
Results, improvements and corrections across the tokenizer, network and app.
GitHub · results/ · live comparison
v2
Latest
v2 benchmark results
Fewer tokens than 12 baselines, Gemma 4 included, on 21 of 21 FLORES+ languages and 24 of 26 NTREX-128.
$ python scripts/premium_v2.py
measured
ntrex
12.8% fewer
flores_plus
13.5% fewer
english
−3 tokens
gemma_4
beaten everywhere
12 baselines · 4 min · results/premium_v2.json
All releases
Read release:
v2 benchmark results
Fewer tokens than 12 baselines, Gemma 4 included, on 21 of 21 FLORES+ languages and 24 of 26 NTREX-128.
Read release:
App preview in Expo Go
The Nzinga app runs in Expo Go from the preview channel. EAS Update ships fixes without a new build.
"runtimeVersion": {
"policy": "sdkVersion"
}
Read release:
Sign-in, invites, consent, earnings
The pilot API: sign-in by one-time code, invites, recordings with consent per speaker, earnings.
Read release:
Twelve languages on FLORES+
Akan and Lingala added. Fewer tokens than 12 other tokenizers on all 12.
aka · Akan
lin · Lingala
+ 10 more languages
Read release:
Gemma 3 extension pack
32,000 African word pieces appended to Gemma 3's tokenizer. Every token keeps its ID.
Read release:
First prototype, corrected
The afro32k claim, 14 of 26, compared premiums. Raw tokens: 7 of 26.
- 14 of 26 (premiums)
+ 7 of 26 (raw tokens)
