# Claude Sonnet 4.5 vs DeepSeek V4 Pro

> DeepSeek V4 Pro is the stronger model overall, scoring 54.2 to 44.0 on the Noometry Index.

- Canonical page: https://noometry.com/compare/claude-sonnet-4-5-vs-deepseek-v4-pro
- Last updated: 2026-10-10
- Shared benchmarks: 42

## Summary

- They share 42 benchmarks with published results for both. Claude Sonnet 4.5 scores higher in 3 categories and DeepSeek V4 Pro in 6 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in math, where DeepSeek V4 Pro leads 64.6 to 31.9.
- The biggest single-benchmark swing is NYT Connections (extended): 37.3% for Claude Sonnet 4.5 and 91.3% for DeepSeek V4 Pro.
- DeepSeek V4 Pro is cheaper at $0.43 / $0.87 per million input/output tokens, against $3 / $15 for Claude Sonnet 4.5.
- DeepSeek V4 Pro has downloadable open weights; the other is API-only.

## Snapshot

| | Claude Sonnet 4.5 | DeepSeek V4 Pro |
|---|---|---|
| Provider | Anthropic | DeepSeek |
| Noometry Index | 44.0 | 54.2 |
| Rank | 76 | 28 |
| Context | 1M | 1M |
| Input $/M | $3 | $0.43 |
| Output $/M | $15 | $0.87 |
| Weights | Proprietary | Open |

## Coding

- Claude Sonnet 4.5: 47.8 (#54)
- DeepSeek V4 Pro: 52.2 (#31)

| Benchmark | Claude Sonnet 4.5 | DeepSeek V4 Pro |
|---|---|---|
| SWE-bench Verified | 71.3% | 77.6% |
| LMArena WebDev | 1393 | 1582 |
| SciCode | 44.7% | 51% |
| WeirdML | 47.7% | 66.2% |
| LMArena Coding | 1489 | 1470 |
| ALE-Bench | 796.15 | 1,403 |
| FrontierCode | — | 28.6% |
| SWE-bench Verified (bash only) | 71.4% | — |
| SWE-bench Multilingual | 67% | — |
| GSO | 14.7% | — |
| AlgoTune | 1.52 | — |

## Agentic & Tool Use

- Claude Sonnet 4.5: 38.1 (#30)
- DeepSeek V4 Pro: 32.7 (#54)

| Benchmark | Claude Sonnet 4.5 | DeepSeek V4 Pro |
|---|---|---|
| Vending-Bench 2 | 3,839 | 3,285 |
| Terminal-Bench | 46.5% | — |
| APEX-Agents | — | 47.3% |
| Berkeley Function Calling Leaderboard | 73.2% | — |
| GDPval | 42.5% | — |
| Remote Labor Index | 2.1% | — |
| τ²-bench Airline | 72% | — |
| τ²-bench Banking | 25.3% | — |
| τ²-bench Retail | 72.4% | — |
| τ²-bench Telecom | 84.9% | — |
| Cybench | 60% | — |
| DeepResearch Bench | 52.6% | — |
| OSWorld | 62.9% | — |
| LMArena Search | 1159 | — |
| METR Time Horizons | 67.4% | — |

## Reasoning

- Claude Sonnet 4.5: 26.9 (#113)
- DeepSeek V4 Pro: 56.4 (#22)

| Benchmark | Claude Sonnet 4.5 | DeepSeek V4 Pro |
|---|---|---|
| ARC-AGI-2 | 13.6% | 61.3% |
| Kagi LLM Benchmark | 57.9% | 53.5% |
| NYT Connections (extended) | 37.3% | 91.3% |
| ARC-AGI-1 | 63.7% | 90.5% |
| CritPt | 1.1% | 18% |
| Chess Puzzles | 12% | 47% |
| LMArena Hard Prompts | 1462 | 1461 |
| Mystery Game Puzzles | 17% | 43% |
| DTBench | 83.2% | 93.9% |
| LMCA | 38.8% | 45.5% |
| Epoch Capabilities Index | 146.84 | 155.39 |
| ForecastBench | 61.9 | 56.1 |
| SimpleBench | 54.3% | — |
| EnigmaEval | 6% | — |
| EBR-Bench | 2.4% | — |
| Surface Evolver Bench | — | 40% |

## Math

- Claude Sonnet 4.5: 31.9 (#190)
- DeepSeek V4 Pro: 64.6 (#24)

| Benchmark | Claude Sonnet 4.5 | DeepSeek V4 Pro |
|---|---|---|
| FrontierMath (Tiers 1-3) | 23.9% | 64.6% |
| FrontierMath Tier 4 | 2.4% | 26.8% |
| OTIS Mock AIME 2024-2025 | 77.8% | 98.6% |
| ProofBench | 19% | 50% |
| LMArena Math | 1448 | 1455 |
| MathArena Final-Answer Competitions | — | 76.6% |
| Omni-MATH | 55.3% | — |
| MATH Level 5 | 97.7% | — |
| FrontierMath (Feb 2025 set) | 15.2% | — |
| FrontierMath Tier 4 (v1) | 4.2% | — |

## Knowledge

- Claude Sonnet 4.5: 48.4 (#70)
- DeepSeek V4 Pro: 59.5 (#28)

| Benchmark | Claude Sonnet 4.5 | DeepSeek V4 Pro |
|---|---|---|
| GPQA Diamond | 82.3% | 91.7% |
| SimpleQA Verified | 30.7% | 52.9% |
| Vectara Hallucination Rate | 12% | 8.6% |
| LMArena Expert | 1483 | 1463 |
| Humanity's Last Exam | 13.7% | — |
| MMLU-Pro | 86.9% | — |
| GPQA (HELM) | 68.6% | — |

## Multimodal

- Claude Sonnet 4.5: 34.1 (#84)
- DeepSeek V4 Pro: —

| Benchmark | Claude Sonnet 4.5 | DeepSeek V4 Pro |
|---|---|---|
| VPCT | 39.8% | — |
| LMArena Document | 1450 | — |

## Multilingual

- Claude Sonnet 4.5: 53.6 (#65)
- DeepSeek V4 Pro: 54.5 (#44)

| Benchmark | Claude Sonnet 4.5 | DeepSeek V4 Pro |
|---|---|---|
| LMArena Non-English | 1425 | 1438 |
| LMArena Chinese | 1459 | 1486 |
| LMArena French | 1459 | 1473 |
| LMArena German | 1427 | 1458 |
| LMArena Japanese | 1390 | 1448 |
| LMArena Korean | 1403 | 1452 |
| LMArena Russian | 1436 | 1452 |
| LMArena Spanish | 1457 | 1457 |

## Instruction Following

- Claude Sonnet 4.5: 75.2 (#74)
- DeepSeek V4 Pro: 76.3 (#44)

| Benchmark | Claude Sonnet 4.5 | DeepSeek V4 Pro |
|---|---|---|
| LMArena Instruction Following | 1459 | 1448 |
| IFEval | 85% | — |

## Long Context

- Claude Sonnet 4.5: 45.2 (#45)
- DeepSeek V4 Pro: 45.0 (#49)

| Benchmark | Claude Sonnet 4.5 | DeepSeek V4 Pro |
|---|---|---|
| LMArena Longer Query | 1476 | 1459 |
| CL-bench Life | — | 13.5% |

## Writing & Preference

- Claude Sonnet 4.5: 66.7 (#33)
- DeepSeek V4 Pro: 65.6 (#45)

| Benchmark | Claude Sonnet 4.5 | DeepSeek V4 Pro |
|---|---|---|
| LMArena Text | 1439 | 1451 |
| LMArena Creative Writing | 1442 | 1446 |
| EQ-Bench Creative Writing | 1678 | 1553 |
| LMArena Multi-Turn | 1465 | 1467 |
| WildBench | 85.4% | — |
| EQ-Bench 4 | — | 1166 |

## FAQ

### Is Claude Sonnet 4.5 better than DeepSeek V4 Pro?

DeepSeek V4 Pro is the stronger model overall, scoring 54.2 to 44.0 on the Noometry Index.

### Which is cheaper, Claude Sonnet 4.5 or DeepSeek V4 Pro?

DeepSeek V4 Pro is cheaper. It lists at $0.43 per million input tokens and $0.87 per million output tokens; Claude Sonnet 4.5 lists at $3 and $15.

### Is Claude Sonnet 4.5 or DeepSeek V4 Pro better for coding?

DeepSeek V4 Pro scores higher on coding benchmarks: 52.2 versus 47.8 in the Noometry coding category.

### Which has the bigger context window?

Both accept 1M tokens.

### How many benchmarks do Claude Sonnet 4.5 and DeepSeek V4 Pro share?

42 benchmarks have published results for both models. Claude Sonnet 4.5 has 73 scored results on Noometry and DeepSeek V4 Pro has 48.