# DeepSeek-R1 vs GPT-5.4 mini

> GPT-5.4 mini is the stronger model overall, scoring 45.0 to 42.3 on the Noometry Index. DeepSeek-R1 costs 1.8× less per token, which makes it the better buy when GPT-5.4 mini's lead doesn't matter for your workload.

- Canonical page: https://noometry.com/compare/deepseek-r1-vs-gpt-5-4-mini
- Last updated: 2026-10-12
- Shared benchmarks: 31

## Summary

- They share 31 benchmarks with published results for both. DeepSeek-R1 scores higher in 4 categories and GPT-5.4 mini in 5 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where GPT-5.4 mini leads 30.4 to 18.6.
- The biggest single-benchmark swing is ARC-AGI-1: 21.2% for DeepSeek-R1 and 63.7% for GPT-5.4 mini.
- DeepSeek-R1 is cheaper at $0.50 / $2.15 per million input/output tokens, against $0.75 / $4.50 for GPT-5.4 mini.
- GPT-5.4 mini accepts more context: 400K tokens versus 164K.

## Snapshot

| | DeepSeek-R1 | GPT-5.4 mini |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 42.3 | 45.0 |
| Rank | 115 | 76 |
| Context | 164K | 400K |
| Input $/M | $0.50 | $0.75 |
| Output $/M | $2.15 | $4.50 |
| Weights | Proprietary | Proprietary |

## Coding

- DeepSeek-R1: 46.3 (#68)
- GPT-5.4 mini: 45.2 (#72)

| Benchmark | DeepSeek-R1 | GPT-5.4 mini |
|---|---|---|
| SciCode | 35.7% | 49.9% |
| WeirdML | 41.6% | 60.3% |
| LMArena Coding | 1427 | 1438 |
| ALE-Bench | 804.12 | 1,189 |
| FrontierCode | — | 27% |
| Aider Polyglot | 71.4% | — |
| LMArena WebDev | — | 1397 |
| LiveBench Coding | 66.7% | — |
| AlgoTune | 1.7 | — |

## Agentic & Tool Use

- DeepSeek-R1: 30.7 (#75)
- GPT-5.4 mini: 29.9 (#81)

| Benchmark | DeepSeek-R1 | GPT-5.4 mini |
|---|---|---|
| DeepResearch Bench | 35.1% | 36.3% |
| BALROG | 34.9% | — |
| METR Time Horizons | 53.8% | — |

## Reasoning

- DeepSeek-R1: 18.6 (#278)
- GPT-5.4 mini: 30.4 (#85)

| Benchmark | DeepSeek-R1 | GPT-5.4 mini |
|---|---|---|
| ARC-AGI-2 | 1.3% | 18.9% |
| Kagi LLM Benchmark | 69.4% | 37.9% |
| ARC-AGI-1 | 21.2% | 63.7% |
| CritPt | 1.1% | 10% |
| LMArena Hard Prompts | 1416 | 1424 |
| Epoch Capabilities Index | 141.29 | 148.84 |
| ForecastBench | 60 | 57 |
| SimpleBench | 40.8% | — |
| NYT Connections (extended) | — | 61.8% |
| Chess Puzzles | — | 24% |
| Thematic Generalization | — | 61.7% |
| LiveBench Reasoning | 83.2% | — |
| Mystery Game Puzzles | — | 11% |
| DTBench | — | 80% |
| LiveBench Data Analysis | 69.8% | — |
| LMCA | — | 40.8% |
| LiveBench | 71.6% | — |

## Math

- DeepSeek-R1: 43.8 (#79)
- GPT-5.4 mini: 45.5 (#75)

| Benchmark | DeepSeek-R1 | GPT-5.4 mini |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 66.4% | 88.9% |
| LMArena Math | 1400 | 1419 |
| FrontierMath (Tiers 1-3) | — | 51.2% |
| FrontierMath Tier 4 | — | 9.8% |
| ProofBench | — | 21% |
| Omni-MATH | 42.4% | — |
| LiveBench Math | 80.7% | — |
| MATH Level 5 | 96.6% | — |
| FrontierMath (Feb 2025 set) | — | 28.3% |
| FrontierMath Tier 4 (v1) | — | 2.1% |

## Knowledge

- DeepSeek-R1: 44.5 (#87)
- GPT-5.4 mini: 51.5 (#67)

| Benchmark | DeepSeek-R1 | GPT-5.4 mini |
|---|---|---|
| GPQA Diamond | 76.3% | 86.9% |
| Vectara Hallucination Rate | 11.3% | 5.5% |
| LMArena Expert | 1394 | 1435 |
| SimpleQA Verified | — | 29.4% |
| MMLU-Pro | 79.3% | — |
| Confabulations | 12.7% | — |
| GPQA (HELM) | 66.6% | — |

## Multimodal

- DeepSeek-R1: —
- GPT-5.4 mini: 39.7 (#56)

| Benchmark | DeepSeek-R1 | GPT-5.4 mini |
|---|---|---|
| LMArena Vision | — | 1245 |

## Multilingual

- DeepSeek-R1: 52.4 (#85)
- GPT-5.4 mini: 51.9 (#96)

| Benchmark | DeepSeek-R1 | GPT-5.4 mini |
|---|---|---|
| LMArena Non-English | 1412 | 1405 |
| LMArena Chinese | 1442 | 1446 |
| LMArena French | 1417 | 1440 |
| LMArena German | 1404 | 1409 |
| LMArena Japanese | 1391 | 1374 |
| LMArena Korean | 1360 | 1368 |
| LMArena Russian | 1423 | 1417 |
| LMArena Spanish | 1411 | 1405 |

## Instruction Following

- DeepSeek-R1: 72.0 (#143)
- GPT-5.4 mini: 74.1 (#102)

| Benchmark | DeepSeek-R1 | GPT-5.4 mini |
|---|---|---|
| LMArena Instruction Following | 1382 | 1405 |
| LiveBench Instruction Following | 80.5% | — |
| IFEval | 78.4% | — |

## Long Context

- DeepSeek-R1: 45.4 (#36)
- GPT-5.4 mini: 43.0 (#112)

| Benchmark | DeepSeek-R1 | GPT-5.4 mini |
|---|---|---|
| LMArena Longer Query | 1391 | 1407 |
| Fiction.LiveBench | 75% | — |

## Writing & Preference

- DeepSeek-R1: 61.4 (#88)
- GPT-5.4 mini: 64.0 (#58)

| Benchmark | DeepSeek-R1 | GPT-5.4 mini |
|---|---|---|
| LMArena Text | 1428 | 1412 |
| LMArena Creative Writing | 1405 | 1370 |
| EQ-Bench Creative Writing | 1500 | 1665 |
| LMArena Multi-Turn | 1405 | 1429 |
| Short-Story Creative Writing | 83% | — |
| WildBench | 82.8% | — |
| LiveBench Language | 48.5% | — |

## FAQ

### Is DeepSeek-R1 better than GPT-5.4 mini?

GPT-5.4 mini is the stronger model overall, scoring 45.0 to 42.3 on the Noometry Index. DeepSeek-R1 costs 1.8× less per token, which makes it the better buy when GPT-5.4 mini's lead doesn't matter for your workload.

### Which is cheaper, DeepSeek-R1 or GPT-5.4 mini?

DeepSeek-R1 is cheaper. It lists at $0.50 per million input tokens and $2.15 per million output tokens; GPT-5.4 mini lists at $0.75 and $4.50.

### Is DeepSeek-R1 or GPT-5.4 mini better for coding?

DeepSeek-R1 scores higher on coding benchmarks: 46.3 versus 45.2 in the Noometry coding category.

### Which has the bigger context window?

GPT-5.4 mini does, with 400K tokens against 164K.

### How many benchmarks do DeepSeek-R1 and GPT-5.4 mini share?

31 benchmarks have published results for both models. DeepSeek-R1 has 52 scored results on Noometry and GPT-5.4 mini has 46.