GPT-5.4 Mini (Medium): benchmark results

GPT-5.4 Mini evaluated at the medium reasoning-effort setting. Provider: OpenAI. Released 2026-03-17. Access: API.

Aggregate Rating 1694 ± 12, rank #379 of 2168 rated models, from 118 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
DexHoldem Agentic Perception - Turn Ownership94.4Field accuracy (%; whose turn it is, all 36 problems; 36 tab100
MageBench S21617Combined 1v1 Elo100
Sonar LLM Leaderboard - Java - Vulnerability Density0.06Vulnerabilities per 1,000 lines of code (lower is better)93.1
AI for Education Visual Maths - Number and Operations72.97Accuracy (%)91.2
HoF-Bench (No Context)57CVEs rediscovered at pass@4 (out of 95), without the generat88.9
AI for Education Visual Maths - Statistics and Probability57.14Accuracy (%)88.5
AA-Omniscience Index - Software Engineering (SWE) - Rust66Omniscience Index86.9
AI for Education Visual Maths - Measurement94.59Accuracy (%)84.5
AA-Omniscience Index - Software Engineering (SWE) - Kotlin32Omniscience Index82.4
OCR-MetaReasoning - Layout Semantics73.6Answer Score (%)82.4
AI for Education Visual Maths - Algebra92.31Accuracy (%)81.8
M-GATE - Round-Trip Translation (Hard Subset)0.76Mean normalized judge score over 29 target languages (0-1; h80.9

Interactive version: theaggregate.ai/model?slug=gpt-5-4-mini-medium · How It Works · Data refreshed daily, snapshot 2026-10-20.