GPT-5.4 Nano (Medium): benchmark results
GPT-5.4 Nano evaluated at the medium reasoning-effort setting. Provider: OpenAI. Released 2026-03-17. Access: API.
Aggregate Rating 1630 ± 17, rank #605 of 2168 rated models, from 77 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Sonar LLM Leaderboard - Java - Vulnerability Density | 0.04 | Vulnerabilities per 1,000 lines of code (lower is better) | 99.3 |
| ChainSWE (Oracle) | 64.8 | Per-bug resolution rate (%; share of the 304 bugs in 100 chr | 83.3 |
| ChainSWE (Sequential with Memory) | 44.1 | Per-bug resolution rate (%; share of the 304 bugs in 100 chr | 83.3 |
| ChainSWE (Sequential) | 45.1 | Per-bug resolution rate (%; share of the 304 bugs in 100 chr | 83.3 |
| AA-Omniscience Hallucination Rate | 51.13 | Hallucination Rate (%) | 78.6 |
| AA Terminal-Bench Hard | 33.33 | Accuracy (%) | 77.7 |
| AA-Omniscience Index - Software Engineering (SWE) - Julia | 8 | Omniscience Index | 77.5 |
| AA IFBench | 64.42 | Accuracy (%) | 75.8 |
| CritPt | 5.1 | Accuracy (imported) | 75.5 |
| LisanBench | 0.09 | Mean Path Length / Current Maximum | 73.9 |
| AA CritPt | 5.14 | Accuracy (%) | 72.8 |
| AA-Omniscience Index - Business | -11.1 | Omniscience Index | 67.2 |
Interactive version: theaggregate.ai/model?slug=gpt-5-4-nano-medium · How It Works · Data refreshed daily, snapshot 2026-10-20.