GPT-4.1 Nano: benchmark results
OpenAI's smallest, fastest, cheapest GPT-4.1 tier, keeping the family's 1M-token context for low-latency classification and autocomplete. Provider: OpenAI. Released 2025-04-14. Access: API.
Aggregate Rating 1516 ± 5, rank #731 of 1656 rated models, from 1959 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| RedBench - HarmBench ASR (Direct Request) | 3.44 | Attack Success Rate (%; HarmBench test behaviors, Llama-Guar | 100 |
| RedBench - HarmBench ASR (Human Jailbreaks) | 0 | Attack Success Rate (%; HarmBench test behaviors, Llama-Guar | 100 |
| RedBench - HarmBench ASR (RainbowPlus) | 6.88 | Attack Success Rate (%; HarmBench test behaviors, Llama-Guar | 100 |
| RedBench - HarmBench ASR (Zero-Shot Attacker) | 0.63 | Attack Success Rate (%; HarmBench test behaviors, Llama-Guar | 100 |
| OpenVLM MMT-Bench - One Shot Detection | 95 | Score (%) | 99 |
| EuroEval Dutch Simplification - Duidelijke Taal | 57.78 | Score (%) | 98.9 |
| OpenVLM MMT-Bench - Furniture Keypoint Detection | 75 | Score (%) | 98.5 |
| OpenVLM Q-Bench - What (In-context Others) | 92.2 | Accuracy (%) | 98.4 |
| OpenVLM MMT-Bench - CIM | 76.7 | Score (%) | 97.3 |
| OpenVLM MMT-Bench - Traffic Light Understanding | 100 | Score (%) | 96.4 |
| HELM VHELM - Bias | 95.74 | Mean win rate | 95.7 |
| HELM VHELM - Pairs | 92.97 | PEM | 95.7 |
Interactive version: theaggregate.ai/model?slug=gpt-4-1-nano · How It Works · Data refreshed daily, snapshot 2026-10-20.