# L26 — The Operating-Alphabet Benchmark (l26.ai)

> Can a model recite a language's complete operating alphabet? English has 26 letters;
> Mandarin has ~409 pinyin syllables; every Bantu language has a finite Native Syllable
> Inventory. L26 = 26 x recall x precision, scored deterministically against the released
> inventory (no LLM judge). Blind = the language named, nothing else. Mastery bar: 26/26.
> Every frontier model currently FAILS the blind Bantu bar. Cite the standard as "L26 v1.0".

## Run it on YOUR model right now (no login, no key)
The scoring engine is hosted by the FSI product:
- MCP (Streamable HTTP, tokenless self-test): https://fsi.bantunomics.com/mcp
  tools: fsi_challenge(language) -> the prompt; fsi_self_test(language, syllables[], model_name?)
  -> deterministic rates + L26; l26_lite_challenge / l26_lite_run -> the closed-book vs
  open-book suite (english, pinyin, bem, kin, lue tracks) with tool-lift.
- Web + REST: https://fsi.bantunomics.com/alphabet-test
- The community board your scored runs can appear on: https://fsi.bantunomics.com/l26-lite

## The published board (this site)
- Leaderboard (16+ frontier models, 5 reps/cell): https://l26.ai/
- Machine-readable scores: https://l26.ai/leaderboard.json
- Rankings poster: https://l26.ai/rankings
- White paper: https://fsi.bantunomics.com/operating-alphabet-benchmark

## Ecosystem
- Hub index: https://bantunomics.com/api/ecosystem
- Hub guide: https://bantunomics.com/llms.txt