Gemma 3 27B vs Phi-4
Gemma 3 27B OPEN WEIGHTS · Google DeepMind, released 2025-03-12
Phi-4 OPEN WEIGHTS · Microsoft Research, released 2024-12-12
“Epoch AI, 'AI Benchmarking Hub'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/benchmarks' [online resource].” · epoch.ai/benchmarks · Licensed CC BY 4.0 · data last verified 2026-09-09 · Methodology · Data Sources
Phi-4 leads on more of the 5 benchmarks where both are measured — Gemma 3 27B on 2, Phi-4 on 3. The remaining 4 of 9 tracked benchmarks cannot be compared: 0 have a result for only one of the two models, 4 for neither. Nothing on this page is estimated to fill those in.
Head-to-head
Measured on 5 of 9 tracked benchmarks for both models — Gemma 3 27B leads on 2, Phi-4 leads on 3.
| Benchmark | Gemma 3 27B | Phi-4 | Leads |
|---|---|---|---|
| Epoch Capabilities Index (ECI) | 130.0 | 130.4 | Phi-4 |
| SWE-bench Verified | Not measured | Not measured | Not measured on both |
| GPQA Diamond | 0.477 | 0.561 | Phi-4 |
| MATH Level 5 | 0.740 | 0.649 | Gemma 3 27B |
| OTIS Mock AIME 2024-2025 | 0.225 | 0.138 | Gemma 3 27B |
| FrontierMath | Not measured | Not measured | Not measured on both |
| FrontierMath Tier 4 | Not measured | Not measured | Not measured on both |
| SimpleQA Verified | Not measured | Not measured | Not measured on both |
| Chess Puzzles | 0.000 | 0.010 | Phi-4 |
What remains uncertain
- 4 tracked benchmarks have no published result for either model.
A missing result means Xentir has not found one published by the source it cites, not that the model performs badly. See Data Sources.