GPT-5.4 Nano: benchmark results

OpenAI's smallest, fastest GPT-5.4 tier. Provider: OpenAI. Released 2026-03-17. Access: API.

Unified ELO 1599 ± 1, rank #270 of 1392 rated models, from 171 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
Honeyval96.9Interaction Length (self-reported)100
MLX Benchmark V2 - MLX Embeddings LoRA86.67Accuracy (%)100
Vectara Hallucination Leaderboard96.9Factual Consistency Rate (%)99
AGC-Bench - humor_transfer0.97Dataset z-score96.3
AGC-Bench - hypobench0.42Dataset z-score95.1
MLX Benchmark V2 - Coding75.76Accuracy (%)95
MLX Benchmark V2 - Conceptual92.31Accuracy (%)90
MLX Benchmark V2 - Very Hard82Accuracy (%)90
RAI-Bench - RAG Robustness (LC Abstention)92Rate (%)89.9
AGC-Bench - pun_eval0.87Dataset z-score87.8
AGC-Bench - thenextchapter0.93Dataset z-score86.6
AGC-Bench - pron_vs_prompt1Dataset z-score86.4

Interactive version: theaggregate.ai/model?slug=gpt-5-4-nano · How It Works · Data refreshed daily, snapshot 2026-09-05.