GPT-5.4 (Medium): benchmark results
GPT-5.4 evaluated at the medium reasoning-effort setting. Provider: OpenAI. Released 2026-03-05. Access: API.
Aggregate Rating 1757 ± 7, rank #218 of 2168 rated models, from 225 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Asuka-Bench (OpenHands) | 90.1 | Cumulative weighted task pass rate (%) after three feedback | 100 |
| Asuka-Bench (OpenHands) - Criteria Pass Rate | 95.1 | Cumulative weighted criteria pass rate (%) after three round | 100 |
| Asuka-Bench (OpenHands) - Project Completion | 52 | Project completion rate (%) under OpenHands: the share of th | 100 |
| CADWorld | 17.5 | Task success rate (%; 200 long-horizon FreeCAD tasks in 11 w | 100 |
| DailyReport (Claude Code) | 2.87 | User preference score (1-4: unhelpful, deficient, acceptable | 100 |
| DailyReport (Claude Code) - Factuality | 81.3 | Factuality (0-1 scaled to %), the share of extracted objecti | 100 |
| DailyReport (Claude Code) - Rationality | 93.3 | Rationality (0-1 scaled to %), rubric scores of 0, 0.5 or 1 | 100 |
| DailyReport (Claude Code) - SubTask Pass | 47.8 | Share of subtasks (0-1 scaled to %) that satisfy every rubri | 100 |
| DailyReport (Search Tools) | 2.89 | User preference score (1-4: unhelpful, deficient, acceptable | 100 |
| DailyReport (Search Tools) - Factuality | 83.5 | Factuality (0-1 scaled to %), the share of extracted objecti | 100 |
| DailyReport (Search Tools) - Rationality | 93 | Rationality (0-1 scaled to %), rubric scores of 0, 0.5 or 1 | 100 |
| DailyReport (Search Tools) - SubTask Pass | 48.4 | Share of subtasks (0-1 scaled to %) that satisfy every rubri | 100 |
Interactive version: theaggregate.ai/model?slug=gpt-5-4-medium · How It Works · Data refreshed daily, snapshot 2026-10-20.