# Deepseek Coder v2 vs Phi-4

> Deepseek Coder v2 is the stronger model overall, scoring 35.9 to 31.2 on the Noometry Index.

- Canonical page: https://noometry.com/compare/deepseek-coder-v2-vs-phi-4
- Last updated: 2026-10-12
- Shared benchmarks: 19

## Summary

- They share 19 benchmarks with published results for both. Deepseek Coder v2 scores higher in 5 categories and Phi-4 in 3 categories; 5 gaps are clear of the uncertainty.
- The widest gap is in math, where Deepseek Coder v2 leads 34.9 to 20.8.

## Snapshot

| | Deepseek Coder v2 | Phi-4 |
|---|---|---|
| Provider | DeepSeek | Microsoft |
| Noometry Index | 35.9 | 31.2 |
| Rank | 220 | 279 |
| Context | — | 128K |
| Input $/M | — | $0.07 |
| Output $/M | — | $0.14 |
| Weights | Open | Open |

## Coding

- Deepseek Coder v2: 38.1 (#183)
- Phi-4: 34.4 (#239)

| Benchmark | Deepseek Coder v2 | Phi-4 |
|---|---|---|
| BigCodeBench Instruct | 48.2% | 45.5% |
| LMArena Coding | 1251 | 1231 |
| BigCodeBench Complete | 59.7% | 55.4% |
| LiveBench Coding | — | 30.7% |
| HumanEval+ | 82.3% | — |
| MBPP+ | 75.1% | — |

## Agentic & Tool Use

- Deepseek Coder v2: —
- Phi-4: 22.8 (#128)

| Benchmark | Deepseek Coder v2 | Phi-4 |
|---|---|---|
| Berkeley Function Calling Leaderboard | — | 28.8% |
| BALROG | — | 11.6% |

## Reasoning

- Deepseek Coder v2: 23.6 (#176)
- Phi-4: 17.7 (#291)

| Benchmark | Deepseek Coder v2 | Phi-4 |
|---|---|---|
| LMArena Hard Prompts | 1207 | 1220 |
| Chess Puzzles | — | 1% |
| LiveBench Reasoning | — | 47.8% |
| LiveBench Data Analysis | — | 45.2% |
| Epoch Capabilities Index | — | 130.42 |
| LiveBench | — | 41.6% |
| WinoGrande | 83.7% | — |

## Math

- Deepseek Coder v2: 34.9 (#190)
- Phi-4: 20.8 (#285)

| Benchmark | Deepseek Coder v2 | Phi-4 |
|---|---|---|
| LMArena Math | 1241 | 1246 |
| OTIS Mock AIME 2024-2025 | — | 13.8% |
| LiveBench Math | — | 42% |
| MATH Level 5 | — | 64.9% |
| GSM8K | 94.5% | — |

## Knowledge

- Deepseek Coder v2: 32.3 (#212)
- Phi-4: 32.6 (#209)

| Benchmark | Deepseek Coder v2 | Phi-4 |
|---|---|---|
| LMArena Expert | 1181 | 1203 |
| GPQA Diamond | — | 56.1% |
| Confabulations | — | 29.4% |
| Vectara Hallucination Rate | — | 3.7% |
| ARC (AI2) Challenge | 64.3% | — |
| MMLU | — | 84.8% |

## Multilingual

- Deepseek Coder v2: 36.3 (#240)
- Phi-4: 37.2 (#237)

| Benchmark | Deepseek Coder v2 | Phi-4 |
|---|---|---|
| LMArena Non-English | 1182 | 1197 |
| LMArena Chinese | 1201 | 1212 |
| LMArena French | 1185 | 1224 |
| LMArena German | 1164 | 1222 |
| LMArena Japanese | 1126 | 1158 |
| LMArena Korean | 1104 | 1151 |
| LMArena Russian | 1188 | 1209 |
| LMArena Spanish | 1153 | 1234 |

## Instruction Following

- Deepseek Coder v2: 61.7 (#242)
- Phi-4: 60.4 (#251)

| Benchmark | Deepseek Coder v2 | Phi-4 |
|---|---|---|
| LMArena Instruction Following | 1180 | 1201 |
| LiveBench Instruction Following | — | 58.4% |

## Long Context

- Deepseek Coder v2: 37.0 (#224)
- Phi-4: 36.9 (#226)

| Benchmark | Deepseek Coder v2 | Phi-4 |
|---|---|---|
| LMArena Longer Query | 1219 | 1217 |

## Writing & Preference

- Deepseek Coder v2: 38.2 (#253)
- Phi-4: 40.5 (#244)

| Benchmark | Deepseek Coder v2 | Phi-4 |
|---|---|---|
| LMArena Text | 1191 | 1217 |
| LMArena Creative Writing | 1120 | 1182 |
| LMArena Multi-Turn | 1177 | 1206 |
| Short-Story Creative Writing | — | 62.6% |
| LiveBench Language | — | 25.6% |

## FAQ

### Is Deepseek Coder v2 better than Phi-4?

Deepseek Coder v2 is the stronger model overall, scoring 35.9 to 31.2 on the Noometry Index.

### Is Deepseek Coder v2 or Phi-4 better for coding?

Deepseek Coder v2 scores higher on coding benchmarks: 38.1 versus 34.4 in the Noometry coding category.

### How many benchmarks do Deepseek Coder v2 and Phi-4 share?

19 benchmarks have published results for both models. Deepseek Coder v2 has 24 scored results on Noometry and Phi-4 has 37.