# Deepseek Coder v2 vs DeepSeek-V3

> DeepSeek-V3 is the stronger model overall, scoring 39.5 to 35.9 on the Noometry Index.

- Canonical page: https://noometry.com/compare/deepseek-coder-v2-vs-deepseek-v3
- Last updated: 2026-10-10
- Shared benchmarks: 23

## Summary

- They share 23 benchmarks with published results for both. Deepseek Coder v2 scores higher in 3 categories and DeepSeek-V3 in 5 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in writing & preference, where DeepSeek-V3 leads 57.4 to 38.2.

## Snapshot

| | Deepseek Coder v2 | DeepSeek-V3 |
|---|---|---|
| Provider | DeepSeek | DeepSeek |
| Noometry Index | 35.9 | 39.5 |
| Rank | 220 | 166 |
| Context | — | 164K |
| Input $/M | — | $0.24 |
| Output $/M | — | $0.90 |
| Weights | Open | Open |

## Coding

- Deepseek Coder v2: 38.1 (#183)
- DeepSeek-V3: 42.3 (#106)

| Benchmark | Deepseek Coder v2 | DeepSeek-V3 |
|---|---|---|
| BigCodeBench Instruct | 48.2% | 50% |
| LMArena Coding | 1251 | 1368 |
| BigCodeBench Complete | 59.7% | 62.2% |
| HumanEval+ | 82.3% | 86.6% |
| MBPP+ | 75.1% | 73% |
| Aider Polyglot | — | 55.1% |
| SciCode | — | 35.8% |
| WeirdML | — | 36.1% |
| LiveBench Coding | — | 70.9% |

## Agentic & Tool Use

- Deepseek Coder v2: —
- DeepSeek-V3: —

| Benchmark | Deepseek Coder v2 | DeepSeek-V3 |
|---|---|---|
| METR Time Horizons | — | 49.6% |

## Reasoning

- Deepseek Coder v2: 23.6 (#176)
- DeepSeek-V3: 20.5 (#236)

| Benchmark | Deepseek Coder v2 | DeepSeek-V3 |
|---|---|---|
| LMArena Hard Prompts | 1207 | 1365 |
| WinoGrande | 83.7% | 85.2% |
| SimpleBench | — | 27.2% |
| Kagi LLM Benchmark | — | 52.3% |
| CritPt | — | 0% |
| LiveBench Reasoning | — | 65.8% |
| DTBench | — | 64.8% |
| LiveBench Data Analysis | — | 60.9% |
| LMCA | — | 15.5% |
| BIG-Bench Hard | — | 87.5% |
| Epoch Capabilities Index | — | 135.94 |
| ForecastBench | — | 59.1 |
| HellaSwag | — | 88.9% |
| LiveBench | — | 66.9% |
| PIQA | — | 84.7% |

## Math

- Deepseek Coder v2: 34.9 (#190)
- DeepSeek-V3: 32.1 (#219)

| Benchmark | Deepseek Coder v2 | DeepSeek-V3 |
|---|---|---|
| LMArena Math | 1241 | 1373 |
| OTIS Mock AIME 2024-2025 | — | 37.8% |
| Omni-MATH | — | 40.3% |
| LiveBench Math | — | 73.5% |
| MATH Level 5 | — | 75.5% |
| FrontierMath (Feb 2025 set) | — | 1.7% |
| GSM8K | 94.5% | — |

## Knowledge

- Deepseek Coder v2: 32.3 (#212)
- DeepSeek-V3: 37.5 (#155)

| Benchmark | Deepseek Coder v2 | DeepSeek-V3 |
|---|---|---|
| LMArena Expert | 1181 | 1351 |
| ARC (AI2) Challenge | 64.3% | 95.3% |
| GPQA Diamond | — | 67.6% |
| MMLU-Pro | — | 72.3% |
| Confabulations | — | 26.1% |
| Vectara Hallucination Rate | — | 6.1% |
| GPQA (HELM) | — | 53.8% |
| MMLU | — | 87.2% |
| TriviaQA | — | 82.9% |

## Multilingual

- Deepseek Coder v2: 36.3 (#240)
- DeepSeek-V3: 48.5 (#143)

| Benchmark | Deepseek Coder v2 | DeepSeek-V3 |
|---|---|---|
| LMArena Non-English | 1182 | 1358 |
| LMArena Chinese | 1201 | 1391 |
| LMArena French | 1185 | 1385 |
| LMArena German | 1164 | 1374 |
| LMArena Japanese | 1126 | 1333 |
| LMArena Korean | 1104 | 1319 |
| LMArena Russian | 1188 | 1373 |
| LMArena Spanish | 1153 | 1358 |

## Instruction Following

- Deepseek Coder v2: 61.7 (#242)
- DeepSeek-V3: 72.8 (#130)

| Benchmark | Deepseek Coder v2 | DeepSeek-V3 |
|---|---|---|
| LMArena Instruction Following | 1180 | 1345 |
| LiveBench Instruction Following | — | 81.5% |
| IFEval | — | 83.2% |

## Long Context

- Deepseek Coder v2: 37.0 (#224)
- DeepSeek-V3: 34.0 (#253)

| Benchmark | Deepseek Coder v2 | DeepSeek-V3 |
|---|---|---|
| LMArena Longer Query | 1219 | 1352 |
| Fiction.LiveBench | — | 50% |

## Writing & Preference

- Deepseek Coder v2: 38.2 (#253)
- DeepSeek-V3: 57.4 (#130)

| Benchmark | Deepseek Coder v2 | DeepSeek-V3 |
|---|---|---|
| LMArena Text | 1191 | 1375 |
| LMArena Creative Writing | 1120 | 1364 |
| LMArena Multi-Turn | 1177 | 1389 |
| Short-Story Creative Writing | — | 77% |
| EQ-Bench Creative Writing | — | 1472 |
| WildBench | — | 83% |
| LiveBench Language | — | 49.1% |

## FAQ

### Is Deepseek Coder v2 better than DeepSeek-V3?

DeepSeek-V3 is the stronger model overall, scoring 39.5 to 35.9 on the Noometry Index.

### Is Deepseek Coder v2 or DeepSeek-V3 better for coding?

DeepSeek-V3 scores higher on coding benchmarks: 42.3 versus 38.1 in the Noometry coding category.

### How many benchmarks do Deepseek Coder v2 and DeepSeek-V3 share?

23 benchmarks have published results for both models. Deepseek Coder v2 has 24 scored results on Noometry and DeepSeek-V3 has 60.