GPT-OSS-120B (Low): benchmark results
GPT-OSS-120B evaluated at the low reasoning-effort setting. Provider: OpenAI. Released 2025-08-05. Access: Open.
Aggregate Rating 1568 ± 9, rank #801 of 2168 rated models, from 207 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| GeoBenchLLM - GeoQuestions1089 Coordinates | 72 | Coordinates accuracy (%) | 100 |
| GeoBenchLLM - SpartUN | 86 | Accuracy (%) | 100 |
| GeoBenchLLM - SpatialEvalLLM | 37 | Accuracy (%) | 100 |
| MedPRESS - Safe Stance Adherence Rate | 47.7 | Safe-stance answers (%; turns 1-5, LLM judge) | 94.7 |
| AA Omniscience - Software Engineering (SWE) - Swift | 60 | Accuracy (%) | 94.4 |
| MedPRESS - Turn of Flip | 2.73 | Pressure turns withstood before first unsafe agreement (0-5) | 89.5 |
| MedPRESS - Unsafe Agreement Rate | 34.9 | Unsafe-agreement answers (%; turns 1-5, LLM judge) | 89.5 |
| Twinkle Eval - MMLU | 88 | Accuracy (%) | 89.3 |
| Multi-LCB - Pass@5 T0.6 - Ruby | 69.6 | Pass@5 (%) | 87.5 |
| GeoBenchLLM - GKMC | 79 | Accuracy (%) | 85.7 |
| GeoBenchLLM - GeoQuestions1089 Yes/No | 73 | Accuracy (%) | 85.7 |
| GeoBenchLLM - GeoSQA | 57 | Accuracy (%) | 85.7 |
Interactive version: theaggregate.ai/model?slug=gpt-oss-120b-low · How It Works · Data refreshed daily, snapshot 2026-10-20.