# DeepSeek-V3 vs GPT-5

> GPT-5 is the stronger model overall, scoring 50.9 to 39.2 on the Noometry Index. DeepSeek-V3 costs 8.5× less per token, which makes it the better buy when GPT-5's lead doesn't matter for your workload.

- Canonical page: https://noometry.com/compare/deepseek-v3-vs-gpt-5
- Last updated: 2026-10-10
- Shared benchmarks: 42

## Summary

- They share 42 benchmarks with published results for both. DeepSeek-V3 scores higher in 0 categories and GPT-5 in 8 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in long context, where GPT-5 leads 69.5 to 34.0.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 37.8% for DeepSeek-V3 and 91.4% for GPT-5.
- DeepSeek-V3 is cheaper at $0.24 / $0.90 per million input/output tokens, against $1.25 / $10 for GPT-5.
- GPT-5 accepts more context: 400K tokens versus 164K.
- DeepSeek-V3 has downloadable open weights; the other is API-only.

## Snapshot

| | DeepSeek-V3 | GPT-5 |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 39.2 | 50.9 |
| Rank | 157 | 37 |
| Context | 164K | 400K |
| Input $/M | $0.24 | $1.25 |
| Output $/M | $0.90 | $10 |
| Weights | Open | Proprietary |

## Coding

- DeepSeek-V3: 41.9 (#99)
- GPT-5: 50.7 (#39)

| Benchmark | DeepSeek-V3 | GPT-5 |
|---|---|---|
| Aider Polyglot | 55.1% | 88% |
| SciCode | 35.8% | 42.9% |
| WeirdML | 36.1% | 60.7% |
| LMArena Coding | 1368 | 1436 |
| SWE-bench Verified | — | 73.6% |
| SWE-bench Verified (bash only) | — | 65% |
| LMArena WebDev | — | 1417 |
| GSO | — | 6.9% |
| BigCodeBench Instruct | 50% | — |
| LiveBench Coding | 70.9% | — |
| BigCodeBench Complete | 62.2% | — |
| ALE-Bench | — | 1,162 |
| AlgoTune | — | 1.67 |
| HumanEval+ | 86.6% | — |
| MBPP+ | 73% | — |

## Agentic & Tool Use

- DeepSeek-V3: —
- GPT-5: 33.0 (#53)

| Benchmark | DeepSeek-V3 | GPT-5 |
|---|---|---|
| METR Time Horizons | 49.6% | 69.6% |
| Terminal-Bench | — | 49.6% |
| GDPval | — | 34.8% |
| Remote Labor Index | — | 1.7% |
| DeepResearch Bench | — | 49.6% |
| BALROG | — | 32.8% |
| LMArena Search | — | 1133 |

## Reasoning

- DeepSeek-V3: 20.0 (#226)
- GPT-5: 38.3 (#55)

| Benchmark | DeepSeek-V3 | GPT-5 |
|---|---|---|
| SimpleBench | 27.2% | 56.7% |
| Kagi LLM Benchmark | 52.3% | 72.7% |
| CritPt | 0% | 12.6% |
| LMArena Hard Prompts | 1365 | 1415 |
| DTBench | 64.8% | 90.7% |
| LMCA | 15.5% | 40% |
| Epoch Capabilities Index | 135.95 | 150 |
| ForecastBench | 59.1 | 61.4 |
| ARC-AGI-2 | — | 9.9% |
| ARC-AGI-1 | — | 65.7% |
| Chess Puzzles | — | 37% |
| EnigmaEval | — | 10.5% |
| EBR-Bench | — | 12.7% |
| LiveBench Reasoning | 65.8% | — |
| Mystery Game Puzzles | — | 23% |
| LiveBench Data Analysis | 60.9% | — |
| BIG-Bench Hard | 87.5% | — |
| HellaSwag | 88.9% | — |
| LiveBench | 66.9% | — |
| PIQA | 84.7% | — |
| WinoGrande | 85.2% | — |

## Math

- DeepSeek-V3: 30.4 (#208)
- GPT-5: 54.6 (#36)

| Benchmark | DeepSeek-V3 | GPT-5 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 37.8% | 91.4% |
| Omni-MATH | 40.3% | 64.7% |
| LMArena Math | 1373 | 1406 |
| MATH Level 5 | 75.5% | 98.1% |
| FrontierMath (Feb 2025 set) | 1.7% | 32.4% |
| FrontierMath (Tiers 1-3) | — | 55.4% |
| FrontierMath Tier 4 | — | 22% |
| ProofBench | — | 18% |
| LiveBench Math | 73.5% | — |
| FrontierMath Tier 4 (v1) | — | 12.5% |

## Knowledge

- DeepSeek-V3: 37.5 (#147)
- GPT-5: 56.5 (#37)

| Benchmark | DeepSeek-V3 | GPT-5 |
|---|---|---|
| GPQA Diamond | 67.6% | 86.2% |
| MMLU-Pro | 72.3% | 86.3% |
| Confabulations | 26.1% | 10.3% |
| Vectara Hallucination Rate | 6.1% | 14.7% |
| GPQA (HELM) | 53.8% | 79.2% |
| LMArena Expert | 1352 | 1420 |
| Humanity's Last Exam | — | 25.3% |
| SimpleQA Verified | — | 50.1% |
| ARC (AI2) Challenge | 95.3% | — |
| MMLU | 87.2% | — |
| TriviaQA | 82.9% | — |

## Multimodal

- DeepSeek-V3: —
- GPT-5: 46.1 (#12)

| Benchmark | DeepSeek-V3 | GPT-5 |
|---|---|---|
| LMArena Vision | — | 1232 |
| GeoBench | — | 81% |
| VPCT | — | 66% |

## Multilingual

- DeepSeek-V3: 48.8 (#138)
- GPT-5: 51.5 (#105)

| Benchmark | DeepSeek-V3 | GPT-5 |
|---|---|---|
| LMArena Non-English | 1358 | 1397 |
| LMArena Chinese | 1390 | 1423 |
| LMArena French | 1386 | 1409 |
| LMArena German | 1374 | 1415 |
| LMArena Japanese | 1333 | 1411 |
| LMArena Korean | 1321 | 1361 |
| LMArena Russian | 1372 | 1406 |
| LMArena Spanish | 1358 | 1398 |

## Instruction Following

- DeepSeek-V3: 73.0 (#125)
- GPT-5: 74.0 (#107)

| Benchmark | DeepSeek-V3 | GPT-5 |
|---|---|---|
| IFEval | 83.2% | 87.5% |
| LMArena Instruction Following | 1345 | 1388 |
| LiveBench Instruction Following | 81.5% | — |

## Long Context

- DeepSeek-V3: 34.0 (#248)
- GPT-5: 69.5 (#2)

| Benchmark | DeepSeek-V3 | GPT-5 |
|---|---|---|
| Fiction.LiveBench | 50% | 97.2% |
| LMArena Longer Query | 1353 | 1399 |

## Writing & Preference

- DeepSeek-V3: 57.6 (#124)
- GPT-5: 63.5 (#63)

| Benchmark | DeepSeek-V3 | GPT-5 |
|---|---|---|
| LMArena Text | 1375 | 1406 |
| LMArena Creative Writing | 1364 | 1365 |
| Short-Story Creative Writing | 77% | 86% |
| EQ-Bench Creative Writing | 1472 | 1627 |
| WildBench | 83% | 85.7% |
| LMArena Multi-Turn | 1389 | 1426 |
| LiveBench Language | 49.1% | — |

## FAQ

### Is DeepSeek-V3 better than GPT-5?

GPT-5 is the stronger model overall, scoring 50.9 to 39.2 on the Noometry Index. DeepSeek-V3 costs 8.5× less per token, which makes it the better buy when GPT-5's lead doesn't matter for your workload.

### Which is cheaper, DeepSeek-V3 or GPT-5?

DeepSeek-V3 is cheaper. It lists at $0.24 per million input tokens and $0.90 per million output tokens; GPT-5 lists at $1.25 and $10.

### Is DeepSeek-V3 or GPT-5 better for coding?

GPT-5 scores higher on coding benchmarks: 50.7 versus 41.9 in the Noometry coding category.

### Which has the bigger context window?

GPT-5 does, with 400K tokens against 164K.

### How many benchmarks do DeepSeek-V3 and GPT-5 share?

42 benchmarks have published results for both models. DeepSeek-V3 has 60 scored results on Noometry and GPT-5 has 69.