GPT-OSS-20B (Low): benchmark results
GPT-OSS-20B evaluated at the low reasoning-effort setting. Provider: OpenAI. Released 2025-08-05. Access: Open.
Aggregate Rating 1504 ± 8, rank #1060 of 2168 rated models, from 463 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| GeoBenchLLM - GeoQuestions1089 Yes/No | 77 | Accuracy (%) | 100 |
| GeoBenchLLM - NY-POI | 40 | Accuracy (%) | 100 |
| GeoBenchLLM - StepGame | 72 | Accuracy (%) | 100 |
| PersonalHomeBench - Multiple Choice (Agentic) | 79.6 | Accuracy (%) on 1,000 multiple-choice household questions (e | 100 |
| EuroEval Dutch Simplification - Duidelijke Taal | 56.94 | Score (%) | 97.7 |
| EuroEval Polish Summarization - PSC | 25.36 | chrF3++ (%) | 96.6 |
| EuroEval Spanish NLU - ScaLA ES | 44.57 | Matthews correlation coefficient (%) | 93.4 |
| SommBench - Food and Wine Pairing | 25 | Food and wine pairing: Matthews correlation coefficient (x 1 | 92.5 |
| EuroEval Portuguese NLU - ScaLA PT | 40.49 | Matthews correlation coefficient (%) | 90 |
| EuroEval Slovak NLU - UNER SK | 63.56 | Micro F1 excluding MISC (%) | 89.7 |
| EuroEval Hungarian NLU - Szeged NER | 73.76 | Micro F1 excluding MISC (%) | 89.6 |
| MedPRESS - Safe Stance Adherence Rate | 32.4 | Safe-stance answers (%; turns 1-5, LLM judge) | 89.5 |
Interactive version: theaggregate.ai/model?slug=gpt-oss-20b-low · How It Works · Data refreshed daily, snapshot 2026-10-20.