# Deepseek Coder v2 vs GPT-3.5-turbo

> Deepseek Coder v2 is the stronger model overall, scoring 35.3 to 23.1 on the Noometry Index.

- Canonical page: https://noometry.com/compare/deepseek-coder-v2-vs-gpt-3-5-turbo
- Last updated: 2026-10-10
- Shared benchmarks: 24

## Summary

- They share 24 benchmarks with published results for both. Deepseek Coder v2 scores higher in 8 categories and GPT-3.5-turbo in 0 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in math, where Deepseek Coder v2 leads 32.3 to 6.0.
- The biggest single-benchmark swing is BigCodeBench Complete: 59.7% for Deepseek Coder v2 and 50.6% for GPT-3.5-turbo.
- Deepseek Coder v2 has downloadable open weights; the other is API-only.

## Snapshot

| | Deepseek Coder v2 | GPT-3.5-turbo |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 35.3 | 23.1 |
| Rank | 217 | 345 |
| Context | — | 16K |
| Input $/M | — | $0.50 |
| Output $/M | — | $1.50 |
| Weights | Open | Proprietary |

## Coding

- Deepseek Coder v2: 37.5 (#178)
- GPT-3.5-turbo: 23.5 (#326)

| Benchmark | Deepseek Coder v2 | GPT-3.5-turbo |
|---|---|---|
| BigCodeBench Instruct | 48.2% | 39.1% |
| LMArena Coding | 1251 | 1137 |
| BigCodeBench Complete | 59.7% | 50.6% |
| HumanEval+ | 82.3% | 70.7% |
| MBPP+ | 75.1% | 69.7% |
| WeirdML | — | 3.5% |

## Agentic & Tool Use

- Deepseek Coder v2: —
- GPT-3.5-turbo: —

| Benchmark | Deepseek Coder v2 | GPT-3.5-turbo |
|---|---|---|
| METR Time Horizons | — | 21.5% |

## Reasoning

- Deepseek Coder v2: 22.6 (#175)
- GPT-3.5-turbo: 13.3 (#328)

| Benchmark | Deepseek Coder v2 | GPT-3.5-turbo |
|---|---|---|
| LMArena Hard Prompts | 1207 | 1108 |
| WinoGrande | 83.7% | 81.6% |
| Chess Puzzles | — | 0% |
| Mystery Game Puzzles | — | 3% |
| DTBench | — | 48.5% |
| LMCA | — | 9.7% |
| Adversarial NLI | — | 58.1% |
| BIG-Bench Hard | — | 61.6% |
| CommonsenseQA 2.0 | — | 57% |
| Epoch Capabilities Index | — | 118.48 |
| ForecastBench | — | 50.4 |

## Math

- Deepseek Coder v2: 32.3 (#188)
- GPT-3.5-turbo: 6.0 (#321)

| Benchmark | Deepseek Coder v2 | GPT-3.5-turbo |
|---|---|---|
| LMArena Math | 1241 | 1142 |
| GSM8K | 94.5% | 57.8% |
| FrontierMath (Tiers 1-3) | — | 0% |
| OTIS Mock AIME 2024-2025 | — | 2.2% |
| MATH Level 5 | — | 15.9% |

## Knowledge

- Deepseek Coder v2: 31.7 (#209)
- GPT-3.5-turbo: 10.0 (#298)

| Benchmark | Deepseek Coder v2 | GPT-3.5-turbo |
|---|---|---|
| LMArena Expert | 1181 | 1070 |
| ARC (AI2) Challenge | 64.3% | 87.4% |
| GPQA Diamond | — | 28% |
| BoolQ | — | 87% |
| MMLU | — | 71.4% |
| OpenBookQA | — | 86% |
| TriviaQA | — | 85.8% |

## Multilingual

- Deepseek Coder v2: 36.4 (#235)
- GPT-3.5-turbo: 31.6 (#253)

| Benchmark | Deepseek Coder v2 | GPT-3.5-turbo |
|---|---|---|
| LMArena Non-English | 1182 | 1108 |
| LMArena Chinese | 1201 | 1075 |
| LMArena French | 1184 | 1117 |
| LMArena German | 1164 | 1090 |
| LMArena Japanese | 1126 | 1043 |
| LMArena Korean | 1104 | 1019 |
| LMArena Russian | 1188 | 1123 |
| LMArena Spanish | 1153 | 1120 |

## Instruction Following

- Deepseek Coder v2: 61.9 (#237)
- GPT-3.5-turbo: 58.1 (#257)

| Benchmark | Deepseek Coder v2 | GPT-3.5-turbo |
|---|---|---|
| LMArena Instruction Following | 1180 | 1119 |

## Long Context

- Deepseek Coder v2: 37.0 (#219)
- GPT-3.5-turbo: 34.0 (#249)

| Benchmark | Deepseek Coder v2 | GPT-3.5-turbo |
|---|---|---|
| LMArena Longer Query | 1219 | 1121 |

## Writing & Preference

- Deepseek Coder v2: 38.4 (#249)
- GPT-3.5-turbo: 25.4 (#300)

| Benchmark | Deepseek Coder v2 | GPT-3.5-turbo |
|---|---|---|
| LMArena Text | 1191 | 1125 |
| LMArena Creative Writing | 1120 | 1092 |
| LMArena Multi-Turn | 1176 | 1117 |
| EQ-Bench Creative Writing | — | 451 |

## FAQ

### Is Deepseek Coder v2 better than GPT-3.5-turbo?

Deepseek Coder v2 is the stronger model overall, scoring 35.3 to 23.1 on the Noometry Index.

### Is Deepseek Coder v2 or GPT-3.5-turbo better for coding?

Deepseek Coder v2 scores higher on coding benchmarks: 37.5 versus 23.5 in the Noometry coding category.

### How many benchmarks do Deepseek Coder v2 and GPT-3.5-turbo share?

24 benchmarks have published results for both models. Deepseek Coder v2 has 24 scored results on Noometry and GPT-3.5-turbo has 44.