PARSE (Persian Zero-Shot) - Factoid Multihop Simple: leaderboard
PARSE (Mozafari, Mousavinasab and Jatowt, University of Innsbruck, 2026) subtype of 600 GPT-4o-generated, human-filtered Persian questions: open multi-hop questions with a single short answer, answered with a Persian-language prompt, zero-shot, on Together AI. The score is string containment of the gold answer, split over any extra answers, from 0 to 1; higher is better.
Metric: Contains match (0-1; answer contains the gold string, split over any extra answers; 600 multi-hop questions; Persian-language prompt, zero-shot, Together AI sampling). Source: arxiv.org. Saturation forecast: Estimated already saturated. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.3 70B Instruct | 0.68 | #1353 |
| 2 | Gemma 2 27B (IT) | 0.64 | #1815 |
| 3 | Mistral Small 3 | 0.59 | #1786 |
| 4 | Qwen 2.5 72B Instruct Turbo | 0.53 | #1168 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=parse-persian-zero-shot-factoid-multihop-simple · How It Works · Data refreshed daily, snapshot 2026-11-12.