BANTUNOMICS L26 · Leaderboard Run the test White paper

The L26 Operating-Alphabet Leaderboard

Self-reported scores out of 26 — how much of each language's operating alphabet a model recites blind (unaided). Mastery is 26/26; the only passing grade is total mastery. Cut by license and origin below.

Open vs Proprietary · mean Bantu-blind L26

Open weights6.0
Proprietary11.0

US vs non-US · mean Bantu-blind L26

United States9.8
Outside the US7.1

The finding

The gap is universal — no category, open or closed, US or not, masters the never-declared Bantu operating alphabets. Every model still FAILS the blind bar.

ModelOriginLicense EnglishBemba
blind
Kinyarwanda
blind
Luvale
blind
Bantu meanGrade
claude-opus-4-8 Anthropic · US proprietary 26.0 21.76 13.8 16.47 17.3 FAIL
Gemini 3.6 Flash Google · US proprietary 26.0 19.1 18.0 13.3 16.8 FAIL
Claude Fable 5 ‡ Anthropic · US proprietary 26.0 8.4 19.2 16.4 14.7 FAIL
Gemini 3.5 Flash Google · US proprietary 26.0 13.4 17.2 13.1 14.6 FAIL
claude-sonnet-5 Anthropic · US proprietary 26.0 12.82 15.13 14.19 14.0 FAIL
Gemini 3 Google · US proprietary 26.0 8.9 21.0 11.8 13.9 FAIL
GPT-5.5 OpenAI · US proprietary 26.0 6.2 19.5 9.7 11.8 FAIL
Claude Opus 4.8 Anthropic · US proprietary 26.0 8.0 13.0 12.4 11.1 FAIL
GLM-5.2 Z.ai / Zhipu · China open 26.0 6.1 13.1 11.1 10.1 FAIL
DeepSeek V4 Pro DeepSeek · China open 26.0 6.6 13.3 10.0 10.0 FAIL
Gemini 3.5 Flash-Lite Google · US proprietary 26.0 7.7 11.1 8.5 9.1 FAIL
DeepSeek V4 DeepSeek · China open 26.0 7.8 10.8 8.3 9.0 FAIL
Grok 4.5 SpaceXAI · US proprietary 26.0 6.0 11.2 9.0 8.7 FAIL
Mistral Large 3 Mistral AI · France proprietary 26.0 4.1 9.4 7.6 7.0 FAIL
Kimi K3 Moonshot AI · China open 26.0 4.1 7.1 9.1 6.8 FAIL
Qwen 3.7 Alibaba · China open 26.0 4.7 15.0 0.3 6.7 FAIL
GPT-5.6 Sol OpenAI · US proprietary 26.0 3.1 8.5 8.6 6.7 FAIL
Nemotron 3 Ultra NVIDIA · US open 26.0 6.1 12.4 0.0 6.2 FAIL
MiniMax M3 MiniMax · China open 26.0 3.9 8.5 4.8 5.7 FAIL
Claude Sonnet 5 Anthropic · US proprietary 26.0 4.7 8.3 2.0 5.0 FAIL
Kimi K2.7 Moonshot AI · China open 26.0 4.3 6.2 4.3 4.9 FAIL
GPT-OSS 120B OpenAI · US open 26.0 3.5 6.5 1.6 3.9 FAIL
Command A+ Cohere · Canada proprietary 26.0 2.9 4.8 2.6 3.4 FAIL
Inkling Thinking Machines Lab · US open 26.0 2.3 6.3 0.3 3.0 FAIL
acme-eval-sim unknown 26.0 1.62 1.6 FAIL
Phi-4-reasoning Microsoft · US open 26.0 0.1 0.3 0.2 0.2 FAIL
218 self-test runs recorded · scores are self-reported model names (unverified) · out of 26 (blind). Run yours at the L26 test.