Model comparison
Phi-4 vs Qwen2.5 72B Instruct
Phi-4 and Qwen2.5 72B Instruct score almost the same on the Noometry Index (31.2 vs 31.9), so choose on price, context window or the category you care about most.
Last verified . 26 shared benchmarks.
Microsoft
Alibaba (Qwen)Summary
- They share 26 benchmarks with published results for both. Phi-4 scores higher in 4 categories and Qwen2.5 72B Instruct in 5 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in writing & preference, where Qwen2.5 72B Instruct leads 46.7 to 40.5.
- The biggest single-benchmark swing is Confabulations: 29.4% for Phi-4 and 19.1% for Qwen2.5 72B Instruct.
- Phi-4 is cheaper at $0.07 / $0.14 per million input/output tokens, against $1.40 / $5.60 for Qwen2.5 72B Instruct.
- Qwen2.5 72B Instruct accepts more context: 131K tokens versus 128K.
Side by side
| Phi-4 | Qwen2.5 72B Instruct | |
|---|---|---|
| Provider | Microsoft | Alibaba (Qwen) |
| Noometry Index | 31.2 | 31.9 |
| Released | 2024-12-11 | 2024-09 |
| Weights | Open | Open |
| Context window | 128K | 131K |
| Max output | 4K | 8K |
| Input $ / M tokens | $0.07 | $1.40 |
| Output $ / M tokens | $0.14 | $5.60 |
| Results tracked | 37 | 43 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Phi-4 leads
Phi-4: 34.4 (#239), Qwen2.5 72B Instruct: 33.2 (#260)
| Benchmark | Phi-4 | Qwen2.5 72B Instruct |
|---|---|---|
| BigCodeBench Instruct | 45.5% | 45.8% |
| LMArena Coding | 1231 | 1292 |
| BigCodeBench Complete | 55.4% | 55.9% |
| WeirdML | — | 16% |
| LiveBench Coding | 30.7% | — |
Agentic & Tool Use Too close to call
Phi-4: 22.8 (#128), Qwen2.5 72B Instruct: 22.1 (#133)
| Benchmark | Phi-4 | Qwen2.5 72B Instruct |
|---|---|---|
| BALROG | 11.6% | 16.2% |
| Berkeley Function Calling Leaderboard | 28.8% | — |
| TheAgentCompany | — | 5.7% |
| METR Time Horizons | — | 35.8% |
Reasoning Qwen2.5 72B Instruct leads
Phi-4: 17.7 (#291), Qwen2.5 72B Instruct: 22.3 (#199)
| Benchmark | Phi-4 | Qwen2.5 72B Instruct |
|---|---|---|
| LMArena Hard Prompts | 1220 | 1271 |
| Epoch Capabilities Index | 130.42 | 129 |
| Chess Puzzles | 1% | — |
| LiveBench Reasoning | 47.8% | — |
| DTBench | — | 62.9% |
| LiveBench Data Analysis | 45.2% | — |
| LMCA | — | 13.4% |
| BIG-Bench Hard | — | 79.8% |
| ForecastBench | — | 57.5 |
| HellaSwag | — | 84.8% |
| LiveBench | 41.6% | — |
| PIQA | — | 82.6% |
| WinoGrande | — | 82.3% |
Math Phi-4 leads
Phi-4: 20.8 (#285), Qwen2.5 72B Instruct: 19.3 (#287)
| Benchmark | Phi-4 | Qwen2.5 72B Instruct |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 13.8% | 8.1% |
| LMArena Math | 1246 | 1283 |
| MATH Level 5 | 64.9% | 63.2% |
| Omni-MATH | — | 33% |
| LiveBench Math | 42% | — |
Knowledge Phi-4 leads
Phi-4: 32.6 (#209), Qwen2.5 72B Instruct: 27.0 (#253)
| Benchmark | Phi-4 | Qwen2.5 72B Instruct |
|---|---|---|
| GPQA Diamond | 56.1% | 49.1% |
| Confabulations | 29.4% | 19.1% |
| LMArena Expert | 1203 | 1245 |
| MMLU | 84.8% | 85.3% |
| MMLU-Pro | — | 63.1% |
| Vectara Hallucination Rate | 3.7% | — |
| GPQA (HELM) | — | 42.6% |
| ARC (AI2) Challenge | — | 94.5% |
| TriviaQA | — | 71.9% |
Multilingual Qwen2.5 72B Instruct leads
Phi-4: 37.2 (#237), Qwen2.5 72B Instruct: 41.0 (#213)
| Benchmark | Phi-4 | Qwen2.5 72B Instruct |
|---|---|---|
| LMArena Non-English | 1197 | 1252 |
| LMArena Chinese | 1212 | 1272 |
| LMArena French | 1224 | 1280 |
| LMArena German | 1222 | 1234 |
| LMArena Japanese | 1158 | 1180 |
| LMArena Korean | 1151 | 1188 |
| LMArena Russian | 1209 | 1264 |
| LMArena Spanish | 1234 | 1256 |
Instruction Following Qwen2.5 72B Instruct leads
Phi-4: 60.4 (#251), Qwen2.5 72B Instruct: 65.5 (#221)
| Benchmark | Phi-4 | Qwen2.5 72B Instruct |
|---|---|---|
| LMArena Instruction Following | 1201 | 1254 |
| LiveBench Instruction Following | 58.4% | — |
| IFEval | — | 80.6% |
Long Context Qwen2.5 72B Instruct leads
Phi-4: 36.9 (#226), Qwen2.5 72B Instruct: 38.9 (#188)
| Benchmark | Phi-4 | Qwen2.5 72B Instruct |
|---|---|---|
| LMArena Longer Query | 1217 | 1282 |
Writing & Preference Qwen2.5 72B Instruct leads
Phi-4: 40.5 (#244), Qwen2.5 72B Instruct: 46.7 (#215)
| Benchmark | Phi-4 | Qwen2.5 72B Instruct |
|---|---|---|
| LMArena Text | 1217 | 1269 |
| LMArena Creative Writing | 1182 | 1221 |
| LMArena Multi-Turn | 1206 | 1272 |
| Short-Story Creative Writing | 62.6% | — |
| WildBench | — | 80.2% |
| LiveBench Language | 25.6% | — |
Frequently asked questions
Is Phi-4 better than Qwen2.5 72B Instruct?
Phi-4 and Qwen2.5 72B Instruct score almost the same on the Noometry Index (31.2 vs 31.9), so choose on price, context window or the category you care about most.
Which is cheaper, Phi-4 or Qwen2.5 72B Instruct?
Phi-4 is cheaper. It lists at $0.07 per million input tokens and $0.14 per million output tokens; Qwen2.5 72B Instruct lists at $1.40 and $5.60.
Is Phi-4 or Qwen2.5 72B Instruct better for coding?
Phi-4 scores higher on coding benchmarks: 34.4 versus 33.2 in the Noometry coding category.
Which has the bigger context window?
Qwen2.5 72B Instruct does, with 131K tokens against 128K.
How many benchmarks do Phi-4 and Qwen2.5 72B Instruct share?
26 benchmarks have published results for both models. Phi-4 has 37 scored results on Noometry and Qwen2.5 72B Instruct has 43.