GPT-5.4 Mini (High): benchmark results
GPT-5.4 Mini evaluated at the high reasoning-effort setting. Provider: OpenAI. Released 2026-03-17. Access: API.
Aggregate Rating 1655 ± 12, rank #505 of 2168 rated models, from 180 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| SteerBench-Work | 91.9 | Mean Trial Accuracy (%) | 94.8 |
| Prim - Discovery | 59.34 | Discovery score (%): how well the model states the problem's | 90.9 |
| SteerBench-Work - Pass^5 | 87.7 | Scenarios Correct in All 5 Trials (%) | 89.7 |
| M-GATE - Round-Trip Translation (Hard Subset) | 0.77 | Mean normalized judge score over 29 target languages (0-1; h | 87 |
| Chatbot Arena (Text - Business, Management & Financial Ops) | 1460 | Arena Score | 85.5 |
| Chatbot Arena (Text - Multi-Turn) | 1466 | Arena Score | 83.6 |
| ALE-Bench | 1188.58 | Performance (Self-Refine x1) (imported) | 82.2 |
| Prim - Generation | 50.55 | Generation accuracy (%): exact match of the model's own zero | 81.8 |
| Chatbot Arena (Text - Non-English) | 1437 | Arena Score | 81.2 |
| Chatbot Arena (Text - Russian) | 1451 | Arena Score | 80.9 |
| ORAgentBench - Easy Feasibility | 62 | Mean feasibility (0-100) on the 32 easy tasks: share of subm | 80.8 |
| Chatbot Arena (Text - Expert) | 1481 | Arena Score | 80.6 |
Interactive version: theaggregate.ai/model?slug=gpt-5-4-mini-high · How It Works · Data refreshed daily, snapshot 2026-10-20.