# DeepSeek V4 Flash vs Step 3

> DeepSeek V4 Flash is the stronger model overall, scoring 53.6 to 40.5 on the Noometry Index.

- Canonical page: https://noometry.com/compare/deepseek-v4-flash-vs-step-3
- Last updated: 2026-10-11
- Shared benchmarks: 16

## Summary

- They share 16 benchmarks with published results for both. DeepSeek V4 Flash scores higher in 8 categories and Step 3 in 0 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where DeepSeek V4 Flash leads 53.7 to 28.4.
- The biggest single-benchmark swing is Kagi LLM Benchmark: 52.2% for DeepSeek V4 Flash and 62.3% for Step 3.

## Snapshot

| | DeepSeek V4 Flash | Step 3 |
|---|---|---|
| Provider | DeepSeek | StepFun |
| Noometry Index | 53.6 | 40.5 |
| Rank | 35 | 149 |
| Context | 1M | — |
| Input $/M | $0.15 | — |
| Output $/M | $0.60 | — |
| Weights | Open | Open |

## Coding

- DeepSeek V4 Flash: 47.9 (#59)
- Step 3: 40.1 (#147)

| Benchmark | DeepSeek V4 Flash | Step 3 |
|---|---|---|
| LMArena Coding | 1457 | 1367 |
| FrontierCode | 18.8% | — |
| LMArena WebDev | 1582 | — |
| SciCode | 49.9% | — |
| WeirdML | 63% | — |
| ALE-Bench | 1,306 | — |

## Reasoning

- DeepSeek V4 Flash: 53.7 (#30)
- Step 3: 28.4 (#105)

| Benchmark | DeepSeek V4 Flash | Step 3 |
|---|---|---|
| Kagi LLM Benchmark | 52.2% | 62.3% |
| LMArena Hard Prompts | 1444 | 1355 |
| ARC-AGI-2 | 61.4% | — |
| SimpleBench | 61.1% | — |
| NYT Connections (extended) | 89.6% | — |
| ARC-AGI-1 | 89% | — |
| CritPt | 16.6% | — |
| Chess Puzzles | 33% | — |
| Mystery Game Puzzles | 34% | — |
| DTBench | 90.9% | — |
| LMCA | 41.7% | — |
| Epoch Capabilities Index | 154.49 | — |

## Math

- DeepSeek V4 Flash: 60.3 (#37)
- Step 3: 37.6 (#148)

| Benchmark | DeepSeek V4 Flash | Step 3 |
|---|---|---|
| LMArena Math | 1427 | 1366 |
| FrontierMath (Tiers 1-3) | 57.5% | — |
| FrontierMath Tier 4 | 24.4% | — |
| MathArena Final-Answer Competitions | 76.5% | — |
| OTIS Mock AIME 2024-2025 | 94.4% | — |
| ProofBench | 56% | — |

## Knowledge

- DeepSeek V4 Flash: 55.4 (#48)
- Step 3: 36.8 (#164)

| Benchmark | DeepSeek V4 Flash | Step 3 |
|---|---|---|
| LMArena Expert | 1441 | 1333 |
| GPQA Diamond | 91% | — |
| SimpleQA Verified | 33.6% | — |

## Multimodal

- DeepSeek V4 Flash: —
- Step 3: 35.5 (#86)

| Benchmark | DeepSeek V4 Flash | Step 3 |
|---|---|---|
| LMArena Vision | — | 1177 |

## Multilingual

- DeepSeek V4 Flash: 53.0 (#72)
- Step 3: 46.3 (#159)

| Benchmark | DeepSeek V4 Flash | Step 3 |
|---|---|---|
| LMArena Non-English | 1420 | 1327 |
| LMArena Chinese | 1468 | 1397 |
| LMArena German | 1418 | 1371 |
| LMArena Korean | 1384 | 1269 |
| LMArena Russian | 1428 | 1331 |
| LMArena Spanish | 1436 | 1371 |
| LMArena French | 1439 | — |
| LMArena Japanese | 1406 | — |

## Instruction Following

- DeepSeek V4 Flash: 74.9 (#81)
- Step 3: 70.4 (#164)

| Benchmark | DeepSeek V4 Flash | Step 3 |
|---|---|---|
| LMArena Instruction Following | 1421 | 1332 |

## Long Context

- DeepSeek V4 Flash: 43.8 (#85)
- Step 3: 40.3 (#157)

| Benchmark | DeepSeek V4 Flash | Step 3 |
|---|---|---|
| LMArena Longer Query | 1434 | 1326 |

## Writing & Preference

- DeepSeek V4 Flash: 63.8 (#61)
- Step 3: 54.3 (#151)

| Benchmark | DeepSeek V4 Flash | Step 3 |
|---|---|---|
| LMArena Text | 1432 | 1350 |
| LMArena Creative Writing | 1403 | 1321 |
| LMArena Multi-Turn | 1449 | 1341 |
| EQ-Bench Creative Writing | 1559 | — |

## FAQ

### Is DeepSeek V4 Flash better than Step 3?

DeepSeek V4 Flash is the stronger model overall, scoring 53.6 to 40.5 on the Noometry Index.

### Is DeepSeek V4 Flash or Step 3 better for coding?

DeepSeek V4 Flash scores higher on coding benchmarks: 47.9 versus 40.1 in the Noometry coding category.

### How many benchmarks do DeepSeek V4 Flash and Step 3 share?

16 benchmarks have published results for both models. DeepSeek V4 Flash has 41 scored results on Noometry and Step 3 has 17.