# DeepSeek V4 Flash vs GPT-5.4 mini

> DeepSeek V4 Flash is the stronger model overall, scoring 53.6 to 45.0 on the Noometry Index.

- Canonical page: https://noometry.com/compare/deepseek-v4-flash-vs-gpt-5-4-mini
- Last updated: 2026-10-11
- Shared benchmarks: 39

## Summary

- They share 39 benchmarks with published results for both. DeepSeek V4 Flash scores higher in 7 categories and GPT-5.4 mini in 1 category; 5 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where DeepSeek V4 Flash leads 53.7 to 30.4.
- The biggest single-benchmark swing is ARC-AGI-2: 61.4% for DeepSeek V4 Flash and 18.9% for GPT-5.4 mini.
- DeepSeek V4 Flash is cheaper at $0.15 / $0.60 per million input/output tokens, against $0.75 / $4.50 for GPT-5.4 mini.
- DeepSeek V4 Flash accepts more context: 1M tokens versus 400K.
- DeepSeek V4 Flash has downloadable open weights; the other is API-only.

## Snapshot

| | DeepSeek V4 Flash | GPT-5.4 mini |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 53.6 | 45.0 |
| Rank | 35 | 76 |
| Context | 1M | 400K |
| Input $/M | $0.15 | $0.75 |
| Output $/M | $0.60 | $4.50 |
| Weights | Open | Proprietary |

## Coding

- DeepSeek V4 Flash: 47.9 (#59)
- GPT-5.4 mini: 45.2 (#72)

| Benchmark | DeepSeek V4 Flash | GPT-5.4 mini |
|---|---|---|
| FrontierCode | 18.8% | 27% |
| LMArena WebDev | 1582 | 1397 |
| SciCode | 49.9% | 49.9% |
| WeirdML | 63% | 60.3% |
| LMArena Coding | 1457 | 1438 |
| ALE-Bench | 1,306 | 1,189 |

## Agentic & Tool Use

- DeepSeek V4 Flash: —
- GPT-5.4 mini: 29.9 (#81)

| Benchmark | DeepSeek V4 Flash | GPT-5.4 mini |
|---|---|---|
| DeepResearch Bench | — | 36.3% |

## Reasoning

- DeepSeek V4 Flash: 53.7 (#30)
- GPT-5.4 mini: 30.4 (#85)

| Benchmark | DeepSeek V4 Flash | GPT-5.4 mini |
|---|---|---|
| ARC-AGI-2 | 61.4% | 18.9% |
| Kagi LLM Benchmark | 52.2% | 37.9% |
| NYT Connections (extended) | 89.6% | 61.8% |
| ARC-AGI-1 | 89% | 63.7% |
| CritPt | 16.6% | 10% |
| Chess Puzzles | 33% | 24% |
| LMArena Hard Prompts | 1444 | 1424 |
| Mystery Game Puzzles | 34% | 11% |
| DTBench | 90.9% | 80% |
| LMCA | 41.7% | 40.8% |
| Epoch Capabilities Index | 154.49 | 148.84 |
| SimpleBench | 61.1% | — |
| Thematic Generalization | — | 61.7% |
| ForecastBench | — | 57 |

## Math

- DeepSeek V4 Flash: 60.3 (#37)
- GPT-5.4 mini: 45.5 (#75)

| Benchmark | DeepSeek V4 Flash | GPT-5.4 mini |
|---|---|---|
| FrontierMath (Tiers 1-3) | 57.5% | 51.2% |
| FrontierMath Tier 4 | 24.4% | 9.8% |
| OTIS Mock AIME 2024-2025 | 94.4% | 88.9% |
| ProofBench | 56% | 21% |
| LMArena Math | 1427 | 1419 |
| MathArena Final-Answer Competitions | 76.5% | — |
| FrontierMath (Feb 2025 set) | — | 28.3% |
| FrontierMath Tier 4 (v1) | — | 2.1% |

## Knowledge

- DeepSeek V4 Flash: 55.4 (#48)
- GPT-5.4 mini: 51.5 (#67)

| Benchmark | DeepSeek V4 Flash | GPT-5.4 mini |
|---|---|---|
| GPQA Diamond | 91% | 86.9% |
| SimpleQA Verified | 33.6% | 29.4% |
| LMArena Expert | 1441 | 1435 |
| Vectara Hallucination Rate | — | 5.5% |

## Multimodal

- DeepSeek V4 Flash: —
- GPT-5.4 mini: 39.7 (#56)

| Benchmark | DeepSeek V4 Flash | GPT-5.4 mini |
|---|---|---|
| LMArena Vision | — | 1245 |

## Multilingual

- DeepSeek V4 Flash: 53.0 (#72)
- GPT-5.4 mini: 51.9 (#96)

| Benchmark | DeepSeek V4 Flash | GPT-5.4 mini |
|---|---|---|
| LMArena Non-English | 1420 | 1405 |
| LMArena Chinese | 1468 | 1446 |
| LMArena French | 1439 | 1440 |
| LMArena German | 1418 | 1409 |
| LMArena Japanese | 1406 | 1374 |
| LMArena Korean | 1384 | 1368 |
| LMArena Russian | 1428 | 1417 |
| LMArena Spanish | 1436 | 1405 |

## Instruction Following

- DeepSeek V4 Flash: 74.9 (#81)
- GPT-5.4 mini: 74.1 (#102)

| Benchmark | DeepSeek V4 Flash | GPT-5.4 mini |
|---|---|---|
| LMArena Instruction Following | 1421 | 1405 |

## Long Context

- DeepSeek V4 Flash: 43.8 (#85)
- GPT-5.4 mini: 43.0 (#112)

| Benchmark | DeepSeek V4 Flash | GPT-5.4 mini |
|---|---|---|
| LMArena Longer Query | 1434 | 1407 |

## Writing & Preference

- DeepSeek V4 Flash: 63.8 (#61)
- GPT-5.4 mini: 64.0 (#58)

| Benchmark | DeepSeek V4 Flash | GPT-5.4 mini |
|---|---|---|
| LMArena Text | 1432 | 1412 |
| LMArena Creative Writing | 1403 | 1370 |
| EQ-Bench Creative Writing | 1559 | 1665 |
| LMArena Multi-Turn | 1449 | 1429 |

## FAQ

### Is DeepSeek V4 Flash better than GPT-5.4 mini?

DeepSeek V4 Flash is the stronger model overall, scoring 53.6 to 45.0 on the Noometry Index.

### Which is cheaper, DeepSeek V4 Flash or GPT-5.4 mini?

DeepSeek V4 Flash is cheaper. It lists at $0.15 per million input tokens and $0.60 per million output tokens; GPT-5.4 mini lists at $0.75 and $4.50.

### Is DeepSeek V4 Flash or GPT-5.4 mini better for coding?

DeepSeek V4 Flash scores higher on coding benchmarks: 47.9 versus 45.2 in the Noometry coding category.

### Which has the bigger context window?

DeepSeek V4 Flash does, with 1M tokens against 400K.

### How many benchmarks do DeepSeek V4 Flash and GPT-5.4 mini share?

39 benchmarks have published results for both models. DeepSeek V4 Flash has 41 scored results on Noometry and GPT-5.4 mini has 46.