# Claude Sonnet 4 vs DeepSeek-V3

> Claude Sonnet 4 is the stronger model overall, scoring 40.8 to 39.5 on the Noometry Index. DeepSeek-V3 costs 15× less per token, which makes it the better buy when Claude Sonnet 4's lead doesn't matter for your workload.

- Canonical page: https://noometry.com/compare/claude-sonnet-4-vs-deepseek-v3
- Last updated: 2026-10-11
- Shared benchmarks: 42

## Summary

- They share 42 benchmarks with published results for both. Claude Sonnet 4 scores higher in 4 categories and DeepSeek-V3 in 4 categories; 6 gaps are clear of the uncertainty.
- The widest gap is in math, where Claude Sonnet 4 leads 43.3 to 32.1.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 71.1% for Claude Sonnet 4 and 37.8% for DeepSeek-V3.
- DeepSeek-V3 is cheaper at $0.24 / $0.90 per million input/output tokens, against $3 / $15 for Claude Sonnet 4.
- Claude Sonnet 4 accepts more context: 200K tokens versus 164K.
- DeepSeek-V3 has downloadable open weights; the other is API-only.

## Snapshot

| | Claude Sonnet 4 | DeepSeek-V3 |
|---|---|---|
| Provider | Anthropic | DeepSeek |
| Noometry Index | 40.8 | 39.5 |
| Rank | 145 | 166 |
| Context | 200K | 164K |
| Input $/M | $3 | $0.24 |
| Output $/M | $15 | $0.90 |
| Weights | Proprietary | Open |

## Coding

- Claude Sonnet 4: 43.5 (#88)
- DeepSeek-V3: 42.3 (#106)

| Benchmark | Claude Sonnet 4 | DeepSeek-V3 |
|---|---|---|
| Aider Polyglot | 61.3% | 55.1% |
| SciCode | 40% | 35.8% |
| WeirdML | 46.1% | 36.1% |
| LMArena Coding | 1414 | 1368 |
| SWE-bench Verified (bash only) | 64.9% | — |
| GSO | 4.9% | — |
| BigCodeBench Instruct | — | 50% |
| LiveBench Coding | — | 70.9% |
| BigCodeBench Complete | — | 62.2% |
| ALE-Bench | 655.35 | — |
| HumanEval+ | — | 86.6% |
| MBPP+ | — | 73% |

## Agentic & Tool Use

- Claude Sonnet 4: 38.5 (#31)
- DeepSeek-V3: —

| Benchmark | Claude Sonnet 4 | DeepSeek-V3 |
|---|---|---|
| METR Time Horizons | 62% | 49.6% |
| TheAgentCompany | 33.1% | — |
| Cybench | 35% | — |
| DeepResearch Bench | 46.6% | — |
| OSWorld | 43.9% | — |

## Reasoning

- Claude Sonnet 4: 22.9 (#187)
- DeepSeek-V3: 20.5 (#236)

| Benchmark | Claude Sonnet 4 | DeepSeek-V3 |
|---|---|---|
| SimpleBench | 45.5% | 27.2% |
| Kagi LLM Benchmark | 73% | 52.3% |
| CritPt | 0.3% | 0% |
| LMArena Hard Prompts | 1372 | 1365 |
| DTBench | 77.1% | 64.8% |
| LMCA | 29% | 15.5% |
| Epoch Capabilities Index | 141.69 | 135.94 |
| ForecastBench | 60.2 | 59.1 |
| ARC-AGI-2 | 5.9% | — |
| ARC-AGI-1 | 40% | — |
| EnigmaEval | 3.1% | — |
| LiveBench Reasoning | — | 65.8% |
| LiveBench Data Analysis | — | 60.9% |
| BIG-Bench Hard | — | 87.5% |
| HellaSwag | — | 88.9% |
| LiveBench | — | 66.9% |
| PIQA | — | 84.7% |
| WinoGrande | — | 85.2% |

## Math

- Claude Sonnet 4: 43.3 (#80)
- DeepSeek-V3: 32.1 (#219)

| Benchmark | Claude Sonnet 4 | DeepSeek-V3 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 71.1% | 37.8% |
| Omni-MATH | 60.2% | 40.3% |
| LMArena Math | 1375 | 1373 |
| MATH Level 5 | 84.4% | 75.5% |
| FrontierMath (Feb 2025 set) | 4.1% | 1.7% |
| LiveBench Math | — | 73.5% |
| FrontierMath Tier 4 (v1) | 0% | — |

## Knowledge

- Claude Sonnet 4: 41.8 (#108)
- DeepSeek-V3: 37.5 (#155)

| Benchmark | Claude Sonnet 4 | DeepSeek-V3 |
|---|---|---|
| GPQA Diamond | 79.2% | 67.6% |
| MMLU-Pro | 84.3% | 72.3% |
| Confabulations | 13.2% | 26.1% |
| Vectara Hallucination Rate | 10.3% | 6.1% |
| GPQA (HELM) | 70.6% | 53.8% |
| LMArena Expert | 1372 | 1351 |
| Humanity's Last Exam | 7.8% | — |
| ARC (AI2) Challenge | — | 95.3% |
| MMLU | — | 87.2% |
| TriviaQA | — | 82.9% |

## Multimodal

- Claude Sonnet 4: 26.2 (#121)
- DeepSeek-V3: —

| Benchmark | Claude Sonnet 4 | DeepSeek-V3 |
|---|---|---|
| LMArena Vision | 1191 | — |
| GeoBench | 37% | — |
| VPCT | 34% | — |
| MindCube | 44.8% | — |

## Multilingual

- Claude Sonnet 4: 46.7 (#156)
- DeepSeek-V3: 48.5 (#143)

| Benchmark | Claude Sonnet 4 | DeepSeek-V3 |
|---|---|---|
| LMArena Non-English | 1333 | 1358 |
| LMArena Chinese | 1350 | 1391 |
| LMArena French | 1363 | 1385 |
| LMArena German | 1331 | 1374 |
| LMArena Japanese | 1302 | 1333 |
| LMArena Korean | 1291 | 1319 |
| LMArena Russian | 1355 | 1373 |
| LMArena Spanish | 1357 | 1358 |

## Instruction Following

- Claude Sonnet 4: 71.7 (#145)
- DeepSeek-V3: 72.8 (#130)

| Benchmark | Claude Sonnet 4 | DeepSeek-V3 |
|---|---|---|
| IFEval | 84% | 83.2% |
| LMArena Instruction Following | 1376 | 1345 |
| LiveBench Instruction Following | — | 81.5% |

## Long Context

- Claude Sonnet 4: 33.7 (#259)
- DeepSeek-V3: 34.0 (#253)

| Benchmark | Claude Sonnet 4 | DeepSeek-V3 |
|---|---|---|
| Fiction.LiveBench | 46.9% | 50% |
| LMArena Longer Query | 1398 | 1352 |

## Writing & Preference

- Claude Sonnet 4: 57.1 (#132)
- DeepSeek-V3: 57.4 (#130)

| Benchmark | Claude Sonnet 4 | DeepSeek-V3 |
|---|---|---|
| LMArena Text | 1351 | 1375 |
| LMArena Creative Writing | 1345 | 1364 |
| Short-Story Creative Writing | 81.4% | 77% |
| EQ-Bench Creative Writing | 1483 | 1472 |
| WildBench | 83.8% | 83% |
| LMArena Multi-Turn | 1376 | 1389 |
| LiveBench Language | — | 49.1% |

## FAQ

### Is Claude Sonnet 4 better than DeepSeek-V3?

Claude Sonnet 4 is the stronger model overall, scoring 40.8 to 39.5 on the Noometry Index. DeepSeek-V3 costs 15× less per token, which makes it the better buy when Claude Sonnet 4's lead doesn't matter for your workload.

### Which is cheaper, Claude Sonnet 4 or DeepSeek-V3?

DeepSeek-V3 is cheaper. It lists at $0.24 per million input tokens and $0.90 per million output tokens; Claude Sonnet 4 lists at $3 and $15.

### Is Claude Sonnet 4 or DeepSeek-V3 better for coding?

Claude Sonnet 4 scores higher on coding benchmarks: 43.5 versus 42.3 in the Noometry coding category.

### Which has the bigger context window?

Claude Sonnet 4 does, with 200K tokens against 164K.

### How many benchmarks do Claude Sonnet 4 and DeepSeek-V3 share?

42 benchmarks have published results for both models. Claude Sonnet 4 has 58 scored results on Noometry and DeepSeek-V3 has 60.