OpenAI
GPT-5.4 benchmark results
20 results on 10 benchmarks from 4 sources.
Also listed as: gpt-5.4-high · gpt-5.4 · gpt-5.4-2026-03-05_medium · GPT-5.4 (medium) · gpt-5.4-2026-03-05_none · GPT-5.4 (none) · gpt-5.4-2026-03-05_high · GPT-5.4 (high) · gpt-5.4-2026-03-05_low · GPT-5.4 (low)
All benchmark results
| Benchmarks | Score | Rank | Setting | Source type | Date |
|---|---|---|---|---|---|
| LMArena Text | 1475 | 34 | Overall, style control on; 64844 votes; 95% CI 1471–1479“gpt-5.4-high” | Official leaderboard | Oct 8, 2026 |
| LMArena Text | 1465 | 53 | Overall, style control on; 67932 votes; 95% CI 1461–1468“gpt-5.4” | Official leaderboard | Oct 8, 2026 |
| LMArena WebDev | 1400 | 89 | Overall; 3734 votes; 95% CI 1390–1411“gpt-5.4” | Official leaderboard | Oct 8, 2026 |
| LMArena Vision | 1284 | 22 | Overall, style control on; 29177 votes; 95% CI 1278–1290“gpt-5.4-high” | Official leaderboard | Oct 2, 2026 |
| LMArena Vision | 1279 | 29 | Overall, style control on; 22828 votes; 95% CI 1272–1285“gpt-5.4” | Official leaderboard | Oct 2, 2026 |
| GPQA Diamond | 93.3% | 17 | Epoch AI run, accuracy; effort: xhigh | Independent run | Mar 6, 2026 |
| GPQA Diamond | 89.9% | 42 | Epoch AI run, accuracy; effort: high | Independent run | Jul 15, 2026 |
| GPQA Diamond | 88.9% | 51 | Epoch AI run, accuracy; effort: medium | Independent run | Jul 15, 2026 |
| GPQA Diamond | 84.8% | 87 | Epoch AI run, accuracy; effort: low | Independent run | Jul 15, 2026 |
| GPQA Diamond | 74.8% | 143 | Epoch AI run, accuracy; effort: none | Independent run | Jul 15, 2026 |
| MMLU-Pro | 87.5% | 9 | Overall accuracy; data source: Self-Reported | Vendor-reported | Mar 11, 2026 |
| SWE-bench Verified | 76.9% | 8 | Epoch AI run (Epoch scaffold); effort: high | Independent run | Mar 6, 2026 |
| AIME (OTIS Mock AIME 2024–2025) | 97.8% | 31 | Epoch AI run, OTIS Mock AIME 2024–2025; effort: high | Independent run | Jul 15, 2026 |
| AIME (OTIS Mock AIME 2024–2025) | 95.6% | 43 | Epoch AI run, OTIS Mock AIME 2024–2025; effort: medium | Independent run | Jul 15, 2026 |
| AIME (OTIS Mock AIME 2024–2025) | 95.3% | 50 | Epoch AI run, OTIS Mock AIME 2024–2025; effort: xhigh | Independent run | Mar 6, 2026 |
| AIME (OTIS Mock AIME 2024–2025) | 84.4% | 95 | Epoch AI run, OTIS Mock AIME 2024–2025; effort: low | Independent run | Jul 15, 2026 |
| AIME (OTIS Mock AIME 2024–2025) | 57.8% | 152 | Epoch AI run, OTIS Mock AIME 2024–2025; effort: none | Independent run | Jul 15, 2026 |
| Humanity's Last Exam | 36.2% | 9 | Compiled by Epoch AI; effort: xhigh | Third-party compilation via Epoch AI | Mar 5, 2026 |
| τ²-bench | 39.4% | 10 | banking_knowledge domain, pass^1; effort: xhigh; submitted by Sierra; v1.0.1 | Official leaderboard | May 6, 2026 |
| Terminal-Bench 2.0 | 81.8% | 5 | Terminal-Bench 2.0; agent: ForgeCode; effort: unknown thinking | Official leaderboard via Epoch AI | Mar 12, 2026 |
FAQ
Where do these GPT-5.4 scores come from?
From Epoch AI Benchmarking Hub, LMArena leaderboard dataset, MMLU-Pro leaderboard (TIGER-Lab), and τ²-bench leaderboard submissions. Each row links to the original leaderboard, dataset or report, and vendor-reported numbers are labelled.