Up to four models
GPT-5.4 Mini vs GPT-5 mini
On the index (GPT-4 = 100), GPT-5.4 Mini scores 147 and GPT-5 mini 140. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
GPT-5.4 Mini
147131–163
WeightsClosed
ReleasedMar 2026
Price in / out$0.75 / $4.5
Context400K
Core tests taken5 of 9
GPT-5 mini
140124–156
WeightsClosed
ReleasedAug 2025
Price in / out$0.25 / $2.0
Context400K
Core tests taken5 of 9
GPT-5.4 Mini leads GPT-5 mini by 7 points, ahead on 3 of 3 shared core tests, at 2.5× the price.
Test by testthird-party results
GPT-5.4 MiniGPT-5 mini
Every result as a table
| Test | GPT-5.4 Mini | GPT-5 mini |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 86.9% | 75.0% |
| GPQA Diamond (Vals AI) | 83.1% | 80.3% |
| MMLU Pro | 84.5% | 82.2% |
| MMMU Pro | 79.3% | 78.9% |
| SimpleQA Verified | 29.4% | 21.6% |
| Maths | ||
| FrontierMath (tiers 1-3) | 51.2% | 46.7% |
| AIME-style maths (OTIS mock) | 88.9% | 86.7% |
| Chess puzzles | 23.0% | 24.0% |
| FrontierMath (2025 set) | 28.3% | 27.2% |
| FrontierMath tier 4 (2025 set) | 2.1% | 6.3% |
| FrontierMath (tier 4) | 9.8% | 12.2% |
| Coding | ||
| LiveCodeBench | 81.5% | 86.6% |
| SWE-bench | 73.0% | 60.8% |
| Vibe Code Bench v1.1 | 48.0% | 14.2% |
| WeirdML | 60.3% | 52.7% |
| Agents | ||
| DTBench | 80.0% | 80.5% |
| LMCA | 40.8% | 34.2% |
| Mystery game puzzles | 11.0% | 10.0% |
| Professional work | ||
| CorpFin v2 | 60.9% | 60.2% |
| MortgageTax | 63.5% | 66.9% |
| TaxEval v2 | 71.2% | 75.2% |
| Novel problems | ||
| ARC-AGI-2 | 18.9% | 4.4% |
| ARC-AGI-1 | 63.7% | 54.3% |
Tests at least two of these models have taken. Bold rows are the core tests.