GPT-5.4
openai/gpt-5.4-2026-03-05
Creator
OpenAIReleased
2026-03-05Weights
not reportedContext
not reportedPrice in / out per Mtok
not reported / not reportedState
rankedHealth summary
11 of 16 scores sit on benchmarks that still discriminate. Every number here has been measured independently. 6 sit on benchmarks with a high risk of test-set contamination.
Epoch Capability Index
156.7 (155.0–158.9)
Epoch AI's Capability Index. A composite score with no benchmark health record behind it, so it is never this site's ranking basis and is always attributed. Published by Epoch AI under CC BY 4.0. fetched 2026-10-19
Scores
| Benchmark | Config | Score | Harness | Provenance | Measured | Health |
|---|---|---|---|---|---|---|
| Knowledge | ||||||
| GPQAnearing saturation | xhigh | 93.3% | not reported | ind. | 2026-03-05 | superseded |
| Humanity's Last Examactive | xhigh | 36.2% | not reported | ind. | 2026-03-05 | clear |
| Coding | ||||||
| Terminal-Benchactive | base | 81.8% | ForgeCode | ind. | 2026-03-05 | contaminated |
| Reasoning | ||||||
| ARC-AGI-2nearing saturation | xhigh | 74.0% | not reported | ind. | 2026-03-05 | superseded |
| On benchmarks that no longer discriminateKept visible on purpose. These are often the numbers a launch announcement leads with. | ||||||
| SWE-bench Verifiedsaturated | high | 76.9% | not reported | ind. | 2026-03-05 | saturatedcontaminated |
| ARC-AGIsaturated | xhigh | 93.7% | not reported | ind. | 2026-03-05 | saturatedcontaminatedsuperseded |