GPT-5 (Low): benchmark results
GPT-5 evaluated at the low reasoning-effort setting. Provider: OpenAI. Released 2025-08-07. Access: API.
Aggregate Rating 1713 ± 10, rank #319 of 2168 rated models, from 144 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| CALRK-Bench - Information Sufficiency Recognition | 75.4 | Accuracy (times 100) on the 279 Type-ISR questions of CALRK- | 100 |
| FinBalance | 46.5 | Exact balance-sheet accuracy (%): share of records whose rep | 100 |
| PointQ-Bench - Defect Diagnosis | 42.32 | Sample-level F1 (%) of the multi-label defect diagnosis (whi | 100 |
| PointQ-Bench - Quality Reasoning | 8.41 | SSFRQ-5D score (0-10): sum of five 0-2 judge ratings (struct | 100 |
| PointQ-Bench - Usability Grading | 50.13 | Macro-F1 (%) of the three-level usability decision, zero-sho | 100 |
| Sand-Glass (Prompted Compression) - Chinese-English In-Bounds Rate | 53.6 | In-bounds rate (% of translations whose target/source syllab | 100 |
| Sand-Glass (Prompted Compression) - Chinese-Spanish In-Bounds Rate | 62.6 | In-bounds rate (% of translations whose target/source syllab | 100 |
| Sand-Glass (Prompted Compression) - Chinese-Spanish Translation Density | 0.27 | BLEU-rho (back-translation BLEU divided by the target/source | 100 |
| VSAS-Bench (Synchronous) | 77.1 | Mean average accuracy (%) over all Present, Cumulative and F | 100 |
| VSAS-Bench (Synchronous) - Present | 83.1 | Mean average accuracy (%) on Present tasks (queries about ev | 100 |
| AA Omniscience - Software Engineering (SWE) - Dart | 44 | Accuracy (%) | 97.8 |
| FlagEval VQA - Geographic Reasoning | 70.2 | Score | 97.8 |
Interactive version: theaggregate.ai/model?slug=gpt-5-low · How It Works · Data refreshed daily, snapshot 2026-10-20.