O4 Mini: benchmark results
OpenAI's efficient o4-series reasoning model for fast math, coding, and visual tasks. Provider: OpenAI. Released 2025-04-16. Access: API.
Aggregate Rating 1707 ± 3, rank #180 of 1656 rated models, from 1197 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ATAD - Blank-based Choice Anomaly | 68.5 | Accuracy (%) on the T3 blank-based choice anomaly task (pick | 100 |
| AgenticRAGTracer (Comparison) - 4-Hop | 35.9 | Exact match (%) on AgenticRAGTracer's 4-hop parallel compari | 100 |
| AgenticRAGTracer (Inference) - 3-Hop | 41.4 | Exact match (%) on AgenticRAGTracer's 3-hop sequential infer | 100 |
| BI-Bench (Python, BI-Agent Tools) | 66.3 | Accuracy (%; 100 end-to-end business-intelligence queries cu | 100 |
| BI-Bench (SQL) | 48.2 | Accuracy (%; 100 end-to-end business-intelligence queries cu | 100 |
| BI-Bench (SQL, BI-Agent Tools) | 61.9 | Accuracy (%; 100 end-to-end business-intelligence queries cu | 100 |
| C2-Faith - Causal Step Localization | 68 | Exact-match localization rate (%): the share of perturbed ch | 100 |
| ConstraintBench - Order Fulfillment | 90 | Feasibility rate (%): share of the domain's 20 tasks whose r | 100 |
| DailyDroid (Multimodal) | 33.33 | Task success rate (%): the cell prints completed tasks out o | 100 |
| DailyDroid (Text-Only) | 29.33 | Task success rate (%): the cell prints completed tasks out o | 100 |
| EuraGovExam - Computer Science | 92.52 | Accuracy (%) on the computer science questions across the fi | 100 |
| EuraGovExam - Physics | 89.63 | Accuracy (%) on the physics questions across the five region | 100 |
Interactive version: theaggregate.ai/model?slug=o4-mini · How It Works · Data refreshed daily, snapshot 2026-10-20.