GPT-5 (Thinking): benchmark results
OpenAI GPT-5 thinking model for more deliberate reasoning and tool-use tasks. Provider: OpenAI. Released 2025-08-07. Access: API.
Aggregate Rating 1737 ± 13, rank #267 of 2168 rated models, from 80 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| KnotBench - PD Code Match | 58 | Accuracy (%) on the 200 D0 items: whether a PD code describe | 100 |
| KnotBench - Reidemeister Move (PD Codes) | 88 | Accuracy (%) on the 100 B0-S items: which Reidemeister move | 100 |
| KnotBench - Same Canonical PD (PD Codes) | 99 | Accuracy (%) on the 100 A3-S items: whether two PD codes are | 100 |
| KnotBench - Same Chirality (Images) | 49 | Accuracy (%) on the 100 A1-I items: whether two diagrams of | 100 |
| OpenAI GPT-5 System Card - BBQ Disambiguated | 0.88 | Accuracy | 100 |
| OpenAI GPT-5 System Card - HealthBench | 67.2 | Score (%) | 100 |
| OpenAI GPT-5 System Card - HealthBench Hard | 46.2 | Score (%) | 100 |
| OpenAI GPT-5 System Card - MMLU Language Bengali | 0.89 | Accuracy | 100 |
| OpenAI GPT-5 System Card - MMLU Language Chinese | 0.9 | Accuracy | 100 |
| OpenAI GPT-5 System Card - MMLU Language Hindi | 0.9 | Accuracy | 100 |
| OpenAI GPT-5 System Card - MMLU Language Indonesian | 0.91 | Accuracy | 100 |
| OpenAI GPT-5 System Card - MMLU Language Japanese | 0.9 | Accuracy | 100 |
Interactive version: theaggregate.ai/model?slug=gpt-5-thinking · How It Works · Data refreshed daily, snapshot 2026-10-20.