# Phi-4 vs Qwen3.5 27B

> Qwen3.5 27B is the stronger model overall, scoring 41.9 to 31.2 on the Noometry Index. Phi-4 costs 9.4× less per token, which makes it the better buy when Qwen3.5 27B's lead doesn't matter for your workload.

- Canonical page: https://noometry.com/compare/phi-4-vs-qwen3-5-27b
- Last updated: 2026-10-10
- Shared benchmarks: 18

## Summary

- They share 18 benchmarks with published results for both. Phi-4 scores higher in 0 categories and Qwen3.5 27B in 8 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in writing & preference, where Qwen3.5 27B leads 59.3 to 40.5.
- The biggest single-benchmark swing is Vectara Hallucination Rate: 3.7% for Phi-4 and 12.1% for Qwen3.5 27B.
- Phi-4 is cheaper at $0.07 / $0.14 per million input/output tokens, against $0.30 / $2.40 for Qwen3.5 27B.
- Qwen3.5 27B accepts more context: 262K tokens versus 128K.

## Snapshot

| | Phi-4 | Qwen3.5 27B |
|---|---|---|
| Provider | Microsoft | Alibaba (Qwen) |
| Noometry Index | 31.2 | 41.9 |
| Rank | 279 | 127 |
| Context | 128K | 262K |
| Input $/M | $0.07 | $0.30 |
| Output $/M | $0.14 | $2.40 |
| Weights | Open | Open |

## Coding

- Phi-4: 34.4 (#239)
- Qwen3.5 27B: 38.9 (#168)

| Benchmark | Phi-4 | Qwen3.5 27B |
|---|---|---|
| LMArena Coding | 1231 | 1427 |
| LMArena WebDev | — | 1358 |
| WeirdML | — | 39.5% |
| BigCodeBench Instruct | 45.5% | — |
| LiveBench Coding | 30.7% | — |
| BigCodeBench Complete | 55.4% | — |
| ALE-Bench | — | 349.45 |

## Agentic & Tool Use

- Phi-4: 22.8 (#128)
- Qwen3.5 27B: —

| Benchmark | Phi-4 | Qwen3.5 27B |
|---|---|---|
| Berkeley Function Calling Leaderboard | 28.8% | — |
| BALROG | 11.6% | — |
| Vending-Bench 2 | — | 201.98 |

## Reasoning

- Phi-4: 17.7 (#291)
- Qwen3.5 27B: 27.5 (#117)

| Benchmark | Phi-4 | Qwen3.5 27B |
|---|---|---|
| LMArena Hard Prompts | 1220 | 1414 |
| NYT Connections (extended) | — | 47.9% |
| Chess Puzzles | 1% | — |
| Thematic Generalization | — | 45.5% |
| LiveBench Reasoning | 47.8% | — |
| DTBench | — | 82.4% |
| LiveBench Data Analysis | 45.2% | — |
| LMCA | — | 34% |
| Epoch Capabilities Index | 130.42 | — |
| LiveBench | 41.6% | — |

## Math

- Phi-4: 20.8 (#285)
- Qwen3.5 27B: 38.8 (#127)

| Benchmark | Phi-4 | Qwen3.5 27B |
|---|---|---|
| LMArena Math | 1246 | 1429 |
| MathArena Final-Answer Competitions | — | 56.7% |
| OTIS Mock AIME 2024-2025 | 13.8% | — |
| LiveBench Math | 42% | — |
| MATH Level 5 | 64.9% | — |

## Knowledge

- Phi-4: 32.6 (#209)
- Qwen3.5 27B: 38.0 (#150)

| Benchmark | Phi-4 | Qwen3.5 27B |
|---|---|---|
| Vectara Hallucination Rate | 3.7% | 12.1% |
| LMArena Expert | 1203 | 1428 |
| GPQA Diamond | 56.1% | — |
| Confabulations | 29.4% | — |
| MMLU | 84.8% | — |

## Multimodal

- Phi-4: —
- Qwen3.5 27B: 39.4 (#59)

| Benchmark | Phi-4 | Qwen3.5 27B |
|---|---|---|
| LMArena Vision | — | 1241 |

## Multilingual

- Phi-4: 37.2 (#237)
- Qwen3.5 27B: 50.8 (#115)

| Benchmark | Phi-4 | Qwen3.5 27B |
|---|---|---|
| LMArena Non-English | 1197 | 1390 |
| LMArena Chinese | 1212 | 1478 |
| LMArena French | 1224 | 1410 |
| LMArena German | 1222 | 1393 |
| LMArena Japanese | 1158 | 1345 |
| LMArena Korean | 1151 | 1358 |
| LMArena Russian | 1209 | 1390 |
| LMArena Spanish | 1234 | 1407 |

## Instruction Following

- Phi-4: 60.4 (#251)
- Qwen3.5 27B: 73.5 (#119)

| Benchmark | Phi-4 | Qwen3.5 27B |
|---|---|---|
| LMArena Instruction Following | 1201 | 1393 |
| LiveBench Instruction Following | 58.4% | — |

## Long Context

- Phi-4: 36.9 (#226)
- Qwen3.5 27B: 43.1 (#106)

| Benchmark | Phi-4 | Qwen3.5 27B |
|---|---|---|
| LMArena Longer Query | 1217 | 1413 |

## Writing & Preference

- Phi-4: 40.5 (#244)
- Qwen3.5 27B: 59.3 (#111)

| Benchmark | Phi-4 | Qwen3.5 27B |
|---|---|---|
| LMArena Text | 1217 | 1409 |
| LMArena Creative Writing | 1182 | 1362 |
| LMArena Multi-Turn | 1206 | 1410 |
| Short-Story Creative Writing | 62.6% | — |
| LiveBench Language | 25.6% | — |

## FAQ

### Is Phi-4 better than Qwen3.5 27B?

Qwen3.5 27B is the stronger model overall, scoring 41.9 to 31.2 on the Noometry Index. Phi-4 costs 9.4× less per token, which makes it the better buy when Qwen3.5 27B's lead doesn't matter for your workload.

### Which is cheaper, Phi-4 or Qwen3.5 27B?

Phi-4 is cheaper. It lists at $0.07 per million input tokens and $0.14 per million output tokens; Qwen3.5 27B lists at $0.30 and $2.40.

### Is Phi-4 or Qwen3.5 27B better for coding?

Qwen3.5 27B scores higher on coding benchmarks: 38.9 versus 34.4 in the Noometry coding category.

### Which has the bigger context window?

Qwen3.5 27B does, with 262K tokens against 128K.

### How many benchmarks do Phi-4 and Qwen3.5 27B share?

18 benchmarks have published results for both models. Phi-4 has 37 scored results on Noometry and Qwen3.5 27B has 28.