Model comparison
Phi-4 vs Qwen2-72B
Phi-4 is the stronger model overall, scoring 31.2 to 30.0 on the Noometry Index.
Last verified . 23 shared benchmarks.
Microsoft
Alibaba (Qwen)Summary
- They share 23 benchmarks with published results for both. Phi-4 scores higher in 5 categories and Qwen2-72B in 4 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in knowledge, where Phi-4 leads 32.6 to 21.2.
- The biggest single-benchmark swing is MATH Level 5: 64.9% for Phi-4 and 39.1% for Qwen2-72B.
Side by side
| Phi-4 | Qwen2-72B | |
|---|---|---|
| Provider | Microsoft | Alibaba (Qwen) |
| Noometry Index | 31.2 | 30.0 |
| Released | 2024-12-11 | 2024-06-07 |
| Weights | Open | Open |
| Context window | 128K | — |
| Max output | 4K | — |
| Input $ / M tokens | $0.07 | — |
| Output $ / M tokens | $0.14 | — |
| Results tracked | 37 | 26 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Phi-4 leads
Phi-4: 34.4 (#239), Qwen2-72B: 29.1 (#310)
| Benchmark | Phi-4 | Qwen2-72B |
|---|---|---|
| BigCodeBench Instruct | 45.5% | 38.5% |
| LMArena Coding | 1231 | 1196 |
| BigCodeBench Complete | 55.4% | 54% |
| WeirdML | — | 11.3% |
| LiveBench Coding | 30.7% | — |
Agentic & Tool Use Phi-4 leads
Phi-4: 22.8 (#128), Qwen2-72B: 17.0 (#146)
| Benchmark | Phi-4 | Qwen2-72B |
|---|---|---|
| Berkeley Function Calling Leaderboard | 28.8% | — |
| TheAgentCompany | — | 1.1% |
| BALROG | 11.6% | — |
| METR Time Horizons | — | 29.9% |
Reasoning Qwen2-72B leads
Phi-4: 17.7 (#291), Qwen2-72B: 23.2 (#181)
| Benchmark | Phi-4 | Qwen2-72B |
|---|---|---|
| LMArena Hard Prompts | 1220 | 1191 |
| Epoch Capabilities Index | 130.42 | 125.28 |
| Chess Puzzles | 1% | — |
| LiveBench Reasoning | 47.8% | — |
| LiveBench Data Analysis | 45.2% | — |
| LiveBench | 41.6% | — |
Math Qwen2-72B leads
Phi-4: 20.8 (#285), Qwen2-72B: 30.2 (#236)
| Benchmark | Phi-4 | Qwen2-72B |
|---|---|---|
| LMArena Math | 1246 | 1235 |
| MATH Level 5 | 64.9% | 39.1% |
| OTIS Mock AIME 2024-2025 | 13.8% | — |
| LiveBench Math | 42% | — |
Knowledge Phi-4 leads
Phi-4: 32.6 (#209), Qwen2-72B: 21.2 (#275)
| Benchmark | Phi-4 | Qwen2-72B |
|---|---|---|
| GPQA Diamond | 56.1% | 40.8% |
| LMArena Expert | 1203 | 1171 |
| MMLU | 84.8% | 82.4% |
| Confabulations | 29.4% | — |
| Vectara Hallucination Rate | 3.7% | — |
Multilingual Phi-4 leads
Phi-4: 37.2 (#237), Qwen2-72B: 35.9 (#244)
| Benchmark | Phi-4 | Qwen2-72B |
|---|---|---|
| LMArena Non-English | 1197 | 1176 |
| LMArena Chinese | 1212 | 1240 |
| LMArena French | 1224 | 1170 |
| LMArena German | 1222 | 1151 |
| LMArena Japanese | 1158 | 1111 |
| LMArena Korean | 1151 | 1083 |
| LMArena Russian | 1209 | 1169 |
| LMArena Spanish | 1234 | 1169 |
Instruction Following Qwen2-72B leads
Phi-4: 60.4 (#251), Qwen2-72B: 61.7 (#241)
| Benchmark | Phi-4 | Qwen2-72B |
|---|---|---|
| LMArena Instruction Following | 1201 | 1181 |
| LiveBench Instruction Following | 58.4% | — |
Long Context Too close to call
Phi-4: 36.9 (#226), Qwen2-72B: 36.1 (#235)
| Benchmark | Phi-4 | Qwen2-72B |
|---|---|---|
| LMArena Longer Query | 1217 | 1192 |
Writing & Preference Too close to call
Phi-4: 40.5 (#244), Qwen2-72B: 40.8 (#241)
| Benchmark | Phi-4 | Qwen2-72B |
|---|---|---|
| LMArena Text | 1217 | 1203 |
| LMArena Creative Writing | 1182 | 1181 |
| LMArena Multi-Turn | 1206 | 1196 |
| Short-Story Creative Writing | 62.6% | — |
| LiveBench Language | 25.6% | — |
Frequently asked questions
Is Phi-4 better than Qwen2-72B?
Phi-4 is the stronger model overall, scoring 31.2 to 30.0 on the Noometry Index.
Is Phi-4 or Qwen2-72B better for coding?
Phi-4 scores higher on coding benchmarks: 34.4 versus 29.1 in the Noometry coding category.
How many benchmarks do Phi-4 and Qwen2-72B share?
23 benchmarks have published results for both models. Phi-4 has 37 scored results on Noometry and Qwen2-72B has 26.