BantuNomics The case L26 Test Leaderboard Benchmark Home

The Alphabet Test

The two-minute test — and it shows you exactly which parts of the alphabet your model is blind to — its native core, its long vowels, its glides, its prenasalized rows. Rates only; the syllables themselves stay licensed. No login.

Already ran it — or want the bigger picture? See how 13 frontier models scored on this exact test. Every one recites English perfectly and fails every operating alphabet it was never handed — then a 4-billion-parameter model, given the standard, closes most of the gap. The Operating-Alphabet Benchmark → (leaderboard, pass/fail report card, and the proof it’s closeable).
Testing your own model? The full L26 test scores English, Mandarin Pinyin (base + toned) and Bantu on one scale (out of 26) — paste your model’s output or bring an API key, blind or scaffolded. Run the L26 Operating-Alphabet Test →

Taking this test as an AI model or agent? Do this — it's one call, no key:

1. Produce the complete syllable list for the language.
2. Score it with a single HTTP GET — make the request with your HTTP / fetch / code tool (this is a request, not a browser click, so tools that can't open constructed URLs still work):

GET http://fsi.bantunomics.com/api/v1/self-test?language=bem&syllables=<your comma-separated list>

CORS is open, no key — you get your score back as JSON (with an l26_about explaining it).
Can't reach us (sandboxed, no network, or no HTTP tool)? Don't stop there — just output the list and have your human paste it in the box below and click Score it. That path always works.
Also: POST the same endpoint · score in place (append &syllables=…&format=json to this page's URL) · MCP /mcp (fsi_challenge, fsi_self_test) · full contract /llms.txt.

Three steps, about two minutes

  1. Copy the prompt below (choose your language first).
  2. Paste it into any model you like — yours, GPT, Gemini, Claude, anything — and copy its answer.
  3. Paste that answer below and hit “Score it.” You get the result — broken out by syllable type.
Prompt for your model:
Produce the complete set of syllables for Bemba (bem). Return a flat list of syllable strings.

Your model doesn’t need to reach us — it just produces the list; you paste it here and hit Score. The breakdown below is rates only, so it leaks no licensed data.

For agents — no browser, no key

Running this with an LLM or agent? Skip the form. Submit in one call — a GET works even from a GET-only tool:

GET "http://fsi.bantunomics.com/api/v1/self-test?language=bem&syllables=<your-syllables, comma-separated>"

Prefer POST? curl -X POST "http://fsi.bantunomics.com/api/v1/self-test" -H "Content-Type: application/json" -d '{"language":"bem","syllables":[<your-syllables>],"model_name":"your-model"}'

Or over MCP (no token needed for the self-test): { "url": "http://fsi.bantunomics.com/mcp" } — tools fsi_challenge, fsi_self_test.

Full machine guide: /llms.txt · /.well-known/mcp.json · API docs