Enkokilish Bench: leaderboard
ScholarXIV Amharic riddle benchmark (እንቆቅልሽ): share of traditional riddles answered correctly (expected answer contained in output) over 222 items.
Metric: Riddles Solved (%). Source: github.com. Saturation forecast: Around July 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 18.02 | #255 |
| 2 | Gemini 3 Flash | 13.06 | #103 |
| 3 | Gemini 2.0 Flash Lite | 7.21 | #455 |
| 4 | Gemini 2.5 Flash Lite | 6.31 | #432 |
| 5 | DeepSeek V3.2 | 2.7 | #210 |
| 6 | Grok 4 Fast (Non-reasoning) | 1.35 | #266 (Grok 4 Fast) |
| 7 | Llama 3.1 8B | 0.9 | #1130 |
| 8 | Nova Lite | 0.9 | #741 |
| 9 | GPT-OSS-20B | 0.45 | #523 |
| 10 | GLM 4.5 Air | 0 | #403 |
| 11 | OLMo-3.1 32B Think | 0 | #636 |
| 12 | Trinity-Mini | 0 | #767 |
| 13 | Devstral Small 2 | 0 | #491 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=enkokilish-bench · How It Works · Data refreshed daily, snapshot 2026-11-05.