GPT-5.4: benchmark results
Supports computer-use workflows and can look up tool definitions as needed, reducing the amount of tool documentation loaded into each request. Provider: OpenAI. Released 2026-03-05. Access: API.
Aggregate Rating 1783 ± 2, rank #79 of 1656 rated models, from 3594 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| 3D-DefectBench - Geometry (Silver Labels) | 0.24 | Macro MCC (Matthews correlation from -1 to 1, macro-averaged | 100 |
| A3S-Bench - Agent-Specific Operators - General Security Score | 4.52 | General Security Score (1-5 judge rating of the full trace; | 100 |
| A3S-Bench - Agent-Specific Operators - RTR@1 | 15.15 | Risk Trigger Rate (%; cases triggered in at least 1 of 3 run | 100 |
| A3S-Bench - Baseline Operators - RTR@1 | 4.2 | Risk Trigger Rate (%; cases triggered in at least 1 of 3 run | 100 |
| AEC-Bench (Nomic Tools) - Cross-Reference Resolution | 77.5 | Mean reward (0-100) from the task verifier on the 51 cross-r | 100 |
| AEC-Bench (Nomic Tools) - Cross-Reference Tracing | 77.1 | Mean reward (0-100) from the task verifier on the 24 cross-r | 100 |
| AEC-Bench (Nomic Tools) - Spec-Drawing Sync | 71.8 | Mean reward (0-100) from the task verifier on the 16 spec-dr | 100 |
| AEC-Bench - Cross-Reference Resolution | 84.3 | Mean reward (0-100) from the task verifier on the 51 cross-r | 100 |
| AEC-Bench - Cross-Reference Tracing | 89.2 | Mean reward (0-100) from the task verifier on the 24 cross-r | 100 |
| AEC-Bench - Sheet Index Consistency | 97.6 | Mean reward (0-100) from the task verifier on the 14 sheet-i | 100 |
| AEC-Bench - Spec-Drawing Sync | 55 | Mean reward (0-100) from the task verifier on the 16 spec-dr | 100 |
| AGC-Bench - Brainstorming | 0.97 | JRT z-score | 100 |
Interactive version: theaggregate.ai/model?slug=gpt-5-4 · How It Works · Data refreshed daily, snapshot 2026-10-20.