Four ways in — from a free public test to a full program subscription. Each is a real step, and you're never locked into climbing them in order.
Prove it free → measure it on your data → license the platform. The paid pilot is 100% creditable toward a subscription, so nothing you spend to evaluate is lost.
See the whole thesis and verify the gap yourself before you talk to anyone.
Confirm the foundation is real on your own models, hands-on, before any commitment.
The whole foundation across the family, plus full depth on the languages you choose — measured on your own held-out data, for one clean go/no-go.
One subscription to the entire living ecosystem — not a dataset licensed once. Everything today, and everything added while you're subscribed.
The four options above, laid out product by product — so you can see exactly what opens up at each tier.
| PublicFree | EvaluationFree · self-serve | Validation Pilot75 days · creditable | Full Annual SubscriptionTalk to us | |
|---|---|---|---|---|
| Best for | Verifying the gap | Scoring your models on the Alphabet Test | Proving value on your own data | Building on the whole platform |
| Languages | Counts only | Bemba + Chewa (Alphabet Test · assigned) 5 languages (L26 Lite · scores only) | 3 in full depth · all 459 FSIs · 10 on L26 Lite | All 459 released, growing |
| The domains — what each plan unlocks, product by product | ||||
| FSI · operating alphabet · foundational | Public test | ✓ Alphabet Test · 2 langs + L26 Lite · 5 langs | ✓ all 459 released | ✓ all released |
| Tone · tonal homographs | Teaser | — | ✓ your langs | ✓ |
| Nouns · aligned matrix | Teaser | — | ✓ your langs | ✓ |
| Verbs · generative engine | Teaser | — | ✓ your langs | ✓ |
| Numbers · numerals & voice | Calculator | — | ✓ your langs | ✓ |
| Health · clinical language | Teaser | — | ✓ your langs | ✓ |
| Grammar · concord matrix | Sample | — | ✓ your langs | ✓ |
| Stories · connected speech | Teaser | — | ✓ your langs | ✓ |
| Consented native audio | — | — | Sample (per language) | ✓ full, uncapped |
| Bulk export | — | — | — | ✓ HuggingFace / Parquet |
| Private data room | — | — | Pilot workspace | ✓ engagement room |
| Term | — | Renewable | 75 days | Annual |
| Price | Free | Free | Talk to us* | Talk to us |
*The Validation Pilot fee is 100% creditable toward the Full Annual Subscription. Enterprise-only; access is provisioned by BantuNomics.
A separate, non-commercial track for grant, philanthropic and impact capital. It funds the public-interest layer — paid native-speaker recording, non-overridable consent and provenance, open standards, and coverage chosen for health and education need rather than commercial demand. Commercial rights stay separately licensed.
No. Each door is independent — start a pilot, or go straight to a subscription, without an evaluation first. An evaluation is welcome, not required.
Better than refundable — it's 100% creditable. Convert to the annual subscription and the entire pilot fee comes straight off the first year.
Evaluation opens the operating alphabet (FSI) for two languages so you can confirm the core is real. The other domains — Tone, Nouns, Verbs, and the rest — open at the Pilot step, on your chosen languages.
Two different things run on one account. The Alphabet Test is scored on your assigned evaluation languages, currently Bemba + Chewa. L26 Lite is a six-track suite covering five languages — English, Mandarin Pinyin (base and toned), Bemba, Kinyarwanda and Luvale. Kinyarwanda and Luvale come free, over and above your assigned pair. The two grade different target sets, so read their scores separately.
Same tracks, lighter run. L26 Lite is the instrument you point at your own model: a single pass per track, closed-book then tool-assisted, free and headless. The full L26 behind l26.ai is the measurement we run across the field — each language evaluated three to five times to assess repeatability, blind and scaffolded. Lite is single-pass and does not measure run-to-run variance; for a repeatability figure, read the board or talk to us. Cite results as L26 v1.0 — the benchmark is young and still iterating.
No — and we're candid about it. It's a subscription to a living platform that keeps growing wider (more languages) and deeper (more per language). You get everything that exists today and everything added while you're subscribed.
Mission-aligned funders — grant, philanthropic, and impact capital — not AI labs. It's an invite-only, non-commercial view of the work.
Check it yourself in a minute: l26.ai shows today's frontier models acing English and failing the same alphabet test on a Bantu language.
One thing worth understanding about the pricing: BantuNomics is a living platform, not a finished dataset — it grows in two directions at once: wider, as we add more of the 500-plus Bantu languages, and deeper, as each language gains more words, recordings, pronunciation, and grammar. That's why the top tier is a subscription: you're licensing something that never stops improving, not buying a file that goes stale.