Model comparison
Phi-4 vs Qwen3.5 397B-A17B
Qwen3.5 397B-A17B is the stronger model overall, scoring 45.7 to 30.9 on the Noometry Index. Phi-4 costs 15× less per token, which makes it the better buy when Qwen3.5 397B-A17B's lead doesn't matter for your workload.
Last verified . 21 shared benchmarks.
Microsoft
Alibaba (Qwen)Summary
- They share 21 benchmarks with published results for both. Phi-4 scores higher in 0 categories and Qwen3.5 397B-A17B in 9 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in math, where Qwen3.5 397B-A17B leads 44.9 to 19.6.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 13.8% for Phi-4 and 88.9% for Qwen3.5 397B-A17B.
- Phi-4 is cheaper at $0.07 / $0.14 per million input/output tokens, against $0.60 / $3.60 for Qwen3.5 397B-A17B.
- Qwen3.5 397B-A17B accepts more context: 262K tokens versus 128K.
Side by side
| Phi-4 | Qwen3.5 397B-A17B | |
|---|---|---|
| Provider | Microsoft | Alibaba (Qwen) |
| Noometry Index | 30.9 | 45.7 |
| Released | 2024-12-11 | 2026-02-01 |
| Weights | Open | Open |
| Context window | 128K | 262K |
| Max output | 4K | 66K |
| Input $ / M tokens | $0.07 | $0.60 |
| Output $ / M tokens | $0.14 | $3.60 |
| Results tracked | 37 | 36 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Qwen3.5 397B-A17B leads
Phi-4: 33.8 (#237), Qwen3.5 397B-A17B: 41.7 (#108)
| Benchmark | Phi-4 | Qwen3.5 397B-A17B |
|---|---|---|
| LMArena Coding | 1231 | 1464 |
| LMArena WebDev | — | 1400 |
| BigCodeBench Instruct | 45.5% | — |
| LiveBench Coding | 30.7% | — |
| BigCodeBench Complete | 55.4% | — |
Agentic & Tool Use Qwen3.5 397B-A17B leads
Phi-4: 22.6 (#120), Qwen3.5 397B-A17B: 33.2 (#51)
| Benchmark | Phi-4 | Qwen3.5 397B-A17B |
|---|---|---|
| APEX-Agents | — | 24.9% |
| Berkeley Function Calling Leaderboard | 28.8% | — |
| τ²-bench Airline | — | 81.5% |
| τ²-bench Banking | — | 9.8% |
| τ²-bench Retail | — | 84.4% |
| τ²-bench Telecom | — | 97.8% |
| BALROG | 11.6% | — |
Reasoning Qwen3.5 397B-A17B leads
Phi-4: 17.0 (#287), Qwen3.5 397B-A17B: 33.6 (#62)
| Benchmark | Phi-4 | Qwen3.5 397B-A17B |
|---|---|---|
| Chess Puzzles | 1% | 13% |
| LMArena Hard Prompts | 1220 | 1448 |
| Epoch Capabilities Index | 130.43 | 146.64 |
| Kagi LLM Benchmark | — | 73.7% |
| NYT Connections (extended) | — | 58.9% |
| Thematic Generalization | — | 65.1% |
| LiveBench Reasoning | 47.8% | — |
| Mystery Game Puzzles | — | 18% |
| DTBench | — | 87.5% |
| LiveBench Data Analysis | 45.2% | — |
| LMCA | — | 37.9% |
| LiveBench | 41.6% | — |
Math Qwen3.5 397B-A17B leads
Phi-4: 19.6 (#279), Qwen3.5 397B-A17B: 44.9 (#64)
| Benchmark | Phi-4 | Qwen3.5 397B-A17B |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 13.8% | 88.9% |
| LMArena Math | 1246 | 1455 |
| FrontierMath (Tiers 1-3) | — | 31.2% |
| LiveBench Math | 42% | — |
| MATH Level 5 | 64.9% | — |
Knowledge Qwen3.5 397B-A17B leads
Phi-4: 32.5 (#204), Qwen3.5 397B-A17B: 53.3 (#53)
| Benchmark | Phi-4 | Qwen3.5 397B-A17B |
|---|---|---|
| GPQA Diamond | 56.1% | 86.4% |
| LMArena Expert | 1203 | 1461 |
| Confabulations | 29.4% | — |
| Vectara Hallucination Rate | 3.7% | — |
| MMLU | 84.8% | — |
Multimodal Not comparable
Phi-4: —, Qwen3.5 397B-A17B: 40.0 (#42)
| Benchmark | Phi-4 | Qwen3.5 397B-A17B |
|---|---|---|
| LMArena Vision | — | 1262 |
Multilingual Qwen3.5 397B-A17B leads
Phi-4: 37.4 (#232), Qwen3.5 397B-A17B: 54.0 (#56)
| Benchmark | Phi-4 | Qwen3.5 397B-A17B |
|---|---|---|
| LMArena Non-English | 1197 | 1431 |
| LMArena Chinese | 1211 | 1501 |
| LMArena French | 1223 | 1461 |
| LMArena German | 1222 | 1447 |
| LMArena Japanese | 1158 | 1427 |
| LMArena Korean | 1151 | 1382 |
| LMArena Russian | 1209 | 1429 |
| LMArena Spanish | 1234 | 1443 |
Instruction Following Qwen3.5 397B-A17B leads
Phi-4: 60.6 (#246), Qwen3.5 397B-A17B: 75.2 (#73)
| Benchmark | Phi-4 | Qwen3.5 397B-A17B |
|---|---|---|
| LMArena Instruction Following | 1201 | 1424 |
| LiveBench Instruction Following | 58.4% | — |
Long Context Qwen3.5 397B-A17B leads
Phi-4: 36.9 (#221), Qwen3.5 397B-A17B: 44.1 (#71)
| Benchmark | Phi-4 | Qwen3.5 397B-A17B |
|---|---|---|
| LMArena Longer Query | 1217 | 1442 |
Writing & Preference Qwen3.5 397B-A17B leads
Phi-4: 40.7 (#239), Qwen3.5 397B-A17B: 62.5 (#74)
| Benchmark | Phi-4 | Qwen3.5 397B-A17B |
|---|---|---|
| LMArena Text | 1217 | 1438 |
| LMArena Creative Writing | 1182 | 1401 |
| LMArena Multi-Turn | 1206 | 1447 |
| Short-Story Creative Writing | 62.6% | — |
| EQ-Bench Creative Writing | — | 1478 |
| LiveBench Language | 25.6% | — |
Frequently asked questions
Is Phi-4 better than Qwen3.5 397B-A17B?
Qwen3.5 397B-A17B is the stronger model overall, scoring 45.7 to 30.9 on the Noometry Index. Phi-4 costs 15× less per token, which makes it the better buy when Qwen3.5 397B-A17B's lead doesn't matter for your workload.
Which is cheaper, Phi-4 or Qwen3.5 397B-A17B?
Phi-4 is cheaper. It lists at $0.07 per million input tokens and $0.14 per million output tokens; Qwen3.5 397B-A17B lists at $0.60 and $3.60.
Is Phi-4 or Qwen3.5 397B-A17B better for coding?
Qwen3.5 397B-A17B scores higher on coding benchmarks: 41.7 versus 33.8 in the Noometry coding category.
Which has the bigger context window?
Qwen3.5 397B-A17B does, with 262K tokens against 128K.
How many benchmarks do Phi-4 and Qwen3.5 397B-A17B share?
21 benchmarks have published results for both models. Phi-4 has 37 scored results on Noometry and Qwen3.5 397B-A17B has 36.