The newest frontier models just took the alphabet test. The gap didn't close. | BantuNomics
GPT-5.6 Sol and Grok 4.5 shipped in July 2026. Within days we scored them on L26, the operating-alphabet benchmark. The newest generation did not close the foundation gap — one flagship scored below its predecessor — and the data says the gap will not close on its own.
Read the article → FSIFull Syllable Inventories are not data, they're infrastructure!
When an AI lab encounters a Full Syllable Inventory for the first time, the question they ask is what is the licensing fee. That is the wrong frame. FSIs are not a dataset you train on. They are the substrate every downstream operation references — closed, enumerable, standardised. Subscribe accordingly.
Read the article → FSIGPT-5.6, Fable 5, Grok 4.5 Compared on Alphabet Tests — and the Results Are In: All Three Failed | BantuNomics
In July 2026 we gave the three newest frontier AI models the simplest test in language AI: recite a language
Read the article → FSIHow a Bantu child learns to read
The syllable is the unit. Not the letter. By age 6 a Bantu child has internalised the syllabary of their language — the same inventory a frontier LLM in 2026 cannot enumerate. One layer off, and the error propagates through every prediction.
Read the article → FSIIt Takes a Village to Raise AGI
There is a well-known African proverb: it takes a village to raise a child. AGI is no different. If it is going to serve all of humanity, it literally takes a global village to raise it.
Read the article → FSIKimi K3, Inkling, GPT-5.6 Sol, Claude Fable 5, Grok 4.5: Five July Flagships, One Alphabet Test — Open and Closed AI Both Failed | BantuNomics
Ten days in July 2026 produced five flagship AI releases — the largest open model ever (Kimi K3), the leading U.S. open model (Inkling), and three closed frontiers (GPT-5.6 Sol, Claude Fable 5, Grok 4.5). We scored them all on the same alphabet test. Open or closed, every one failed — and the open-vs-closed gap turns out to be a rounding error next to the real divide.
Read the article → FSIThe alphabet test every frontier model fails — L26 Operating-Alphabet Benchmark | BantuNomics
L26 asks a basic question: can a model list every building block a language is made from — its complete alphabet? Frontier models nail English but fail the Bantu alphabets they
Read the article → FSIThe alphabet test for AGI
Imagine the AGI did not know the 26 letters. It would still write fluently — and still fail to sound out a new word, spell, or teach a child to read. You would stop calling it AGI on the spot. That is the model's literal condition today for 700+ Bantu languages: it does not have the alphabet, and it cannot feel that it's missing.
Read the article → FSIThe Bantu Syllable Equation
Every Bantu syllable fits one equation — σ → N₀C₀H₀G₀V. We have now tested it across 700+ languages, and it still holds, with no structural change. That law is why a Full Syllable Inventory takes the shape it does, why validating the parts settles the whole, and why a closed native inventory can tell you which words are borrowed.
Read the article → FSIThe Flat Text Problem
Bantu languages have been spoken for thousands of years. The writing system is barely a hundred years old — and it leaves out the part that decides meaning. We named this gap so the AI labs could see it.
Read the article → FSIThe known unknown
AI labs have a name for it: jagged intelligence. The polite term for the blindness is unknown unknowns. The Bantu family used to be one of those. It isn't anymore — we have named it, mapped it, and measured the gap. And we are starting with the most foundational manifestation: the syllable layer every higher-layer Bantu AI failure mode stacks on top of.
Read the article → FSITone lives in the syllable
Tone in a Bantu language is not decoration — it is meaning. And tone does not attach to the letter or the morpheme. It attaches to the syllable. If a frontier model cannot enumerate the syllables of a Bantu language, it cannot understand its tone. The two failures are the same failure.
Read the article → FSIWhat the missing FSI costs you
You've seen the model can't detect its own missing foundation. Here's what that hole costs you, in numbers you already track: tokens you shouldn't need, products that underperform in the markets you paid most to enter, and an entire error class your text-based evals cannot see. Against any one of them, the subscription is small.
Read the article →