# DeepSeek-V2.5 (Sep 2024) vs Step 3

> Step 3 is the stronger model overall, scoring 40.5 to 37.6 on the Noometry Index.

- Canonical page: https://noometry.com/compare/deepseek-v2-5-vs-step-3
- Last updated: 2026-10-11
- Shared benchmarks: 15

## Summary

- They share 15 benchmarks with published results for both. DeepSeek-V2.5 (Sep 2024) scores higher in 0 categories and Step 3 in 8 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in coding, where Step 3 leads 40.1 to 31.7.

## Snapshot

| | DeepSeek-V2.5 (Sep 2024) | Step 3 |
|---|---|---|
| Provider | DeepSeek | StepFun |
| Noometry Index | 37.6 | 40.5 |
| Rank | 200 | 149 |
| Context | — | — |
| Input $/M | — | — |
| Output $/M | — | — |
| Weights | Open | Open |

## Coding

- DeepSeek-V2.5 (Sep 2024): 31.7 (#281)
- Step 3: 40.1 (#147)

| Benchmark | DeepSeek-V2.5 (Sep 2024) | Step 3 |
|---|---|---|
| LMArena Coding | 1309 | 1367 |
| Aider Polyglot | 17.8% | — |
| BigCodeBench Instruct | 48.6% | — |
| BigCodeBench Complete | 53.2% | — |
| HumanEval+ | 83.5% | — |
| MBPP+ | 74.1% | — |

## Reasoning

- DeepSeek-V2.5 (Sep 2024): 25.6 (#145)
- Step 3: 28.4 (#105)

| Benchmark | DeepSeek-V2.5 (Sep 2024) | Step 3 |
|---|---|---|
| LMArena Hard Prompts | 1289 | 1355 |
| Kagi LLM Benchmark | — | 62.3% |

## Math

- DeepSeek-V2.5 (Sep 2024): 35.9 (#177)
- Step 3: 37.6 (#148)

| Benchmark | DeepSeek-V2.5 (Sep 2024) | Step 3 |
|---|---|---|
| LMArena Math | 1288 | 1366 |

## Knowledge

- DeepSeek-V2.5 (Sep 2024): 34.8 (#193)
- Step 3: 36.8 (#164)

| Benchmark | DeepSeek-V2.5 (Sep 2024) | Step 3 |
|---|---|---|
| LMArena Expert | 1266 | 1333 |

## Multimodal

- DeepSeek-V2.5 (Sep 2024): —
- Step 3: 35.5 (#86)

| Benchmark | DeepSeek-V2.5 (Sep 2024) | Step 3 |
|---|---|---|
| LMArena Vision | — | 1177 |

## Multilingual

- DeepSeek-V2.5 (Sep 2024): 42.5 (#193)
- Step 3: 46.3 (#159)

| Benchmark | DeepSeek-V2.5 (Sep 2024) | Step 3 |
|---|---|---|
| LMArena Non-English | 1273 | 1327 |
| LMArena Chinese | 1318 | 1397 |
| LMArena German | 1258 | 1371 |
| LMArena Korean | 1209 | 1269 |
| LMArena Russian | 1289 | 1331 |
| LMArena Spanish | 1248 | 1371 |
| LMArena French | 1289 | — |
| LMArena Japanese | 1228 | — |

## Instruction Following

- DeepSeek-V2.5 (Sep 2024): 67.5 (#194)
- Step 3: 70.4 (#164)

| Benchmark | DeepSeek-V2.5 (Sep 2024) | Step 3 |
|---|---|---|
| LMArena Instruction Following | 1280 | 1332 |

## Long Context

- DeepSeek-V2.5 (Sep 2024): 39.5 (#174)
- Step 3: 40.3 (#157)

| Benchmark | DeepSeek-V2.5 (Sep 2024) | Step 3 |
|---|---|---|
| LMArena Longer Query | 1301 | 1326 |

## Writing & Preference

- DeepSeek-V2.5 (Sep 2024): 49.8 (#187)
- Step 3: 54.3 (#151)

| Benchmark | DeepSeek-V2.5 (Sep 2024) | Step 3 |
|---|---|---|
| LMArena Text | 1294 | 1350 |
| LMArena Creative Writing | 1285 | 1321 |
| LMArena Multi-Turn | 1297 | 1341 |

## FAQ

### Is DeepSeek-V2.5 (Sep 2024) better than Step 3?

Step 3 is the stronger model overall, scoring 40.5 to 37.6 on the Noometry Index.

### Is DeepSeek-V2.5 (Sep 2024) or Step 3 better for coding?

Step 3 scores higher on coding benchmarks: 40.1 versus 31.7 in the Noometry coding category.

### How many benchmarks do DeepSeek-V2.5 (Sep 2024) and Step 3 share?

15 benchmarks have published results for both models. DeepSeek-V2.5 (Sep 2024) has 22 scored results on Noometry and Step 3 has 17.