IyawoBench (v1.0 Protocol): leaderboard
IyawoBench (Iyawo, 2026) asks for a three-way triage decision (refer now, refer today, treat here) on 200 synthetic febrile-illness vignettes derived from encounters at Nigerian primary health centres. This board holds the original v1.0 paper runs, one user-turn prompt at temperature 0.1 with at most 150 output tokens, scored by triage accuracy in percent with unparseable answers counted wrong; higher is better.
Metric: Triage accuracy (%): share of the 200 IyawoBench v1.0 vignettes (100 refer now, 60 refer today, 40 treat here) whose first parsed triage token (REFER_NOW, REFER_TODAY or TREAT_HERE) matches the author-assigned level; a response without a token counts as wrong; one user-turn prompt citing WHO IMCI 2014, WHO malaria 2025, Surviving Sepsis 2021 and Nigerian treatment guidelines, temperature 0.1, at most 150 output tokens; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 4 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 67.5 | #261 |
| 2 | Llama 4 Scout Instruct | 59.5 | #1579 |
| 3 | Llama 3.3 70B Instruct | 43 | #1385 |
| 4 | Llama 3.1 8B Instruct | 39 | #4600 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=iyawobench-v1-0-protocol · How It Works · Data refreshed daily, snapshot 2026-11-21.