# Claude Opus 4.6 vs DeepSeek-V3

> Claude Opus 4.6 is the stronger model overall, scoring 58.3 to 39.2 on the Noometry Index. DeepSeek-V3 costs 25× less per token, which makes it the better buy when Claude Opus 4.6's lead doesn't matter for your workload.

- Canonical page: https://noometry.com/compare/claude-opus-4-6-vs-deepseek-v3
- Last updated: 2026-10-10
- Shared benchmarks: 30

## Summary

- They share 30 benchmarks with published results for both. Claude Opus 4.6 scores higher in 8 categories and DeepSeek-V3 in 0 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where Claude Opus 4.6 leads 57.7 to 20.0.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 94.4% for Claude Opus 4.6 and 37.8% for DeepSeek-V3.
- DeepSeek-V3 is cheaper at $0.24 / $0.90 per million input/output tokens, against $5 / $25 for Claude Opus 4.6.
- Claude Opus 4.6 accepts more context: 1M tokens versus 164K.
- DeepSeek-V3 has downloadable open weights; the other is API-only.

## Snapshot

| | Claude Opus 4.6 | DeepSeek-V3 |
|---|---|---|
| Provider | Anthropic | DeepSeek |
| Noometry Index | 58.3 | 39.2 |
| Rank | 18 | 157 |
| Context | 1M | 164K |
| Input $/M | $5 | $0.24 |
| Output $/M | $25 | $0.90 |
| Weights | Proprietary | Open |

## Coding

- Claude Opus 4.6: 57.7 (#15)
- DeepSeek-V3: 41.9 (#99)

| Benchmark | Claude Opus 4.6 | DeepSeek-V3 |
|---|---|---|
| WeirdML | 78% | 36.1% |
| LMArena Coding | 1535 | 1368 |
| SWE-bench Verified | 78.7% | — |
| FrontierCode | 26.6% | — |
| SWE-bench Verified (bash only) | 75.6% | — |
| Aider Polyglot | — | 55.1% |
| LMArena WebDev | 1546 | — |
| SWE-bench Multilingual | 72% | — |
| SciCode | — | 35.8% |
| GSO | 41.2% | — |
| BigCodeBench Instruct | — | 50% |
| LiveBench Coding | — | 70.9% |
| BigCodeBench Complete | — | 62.2% |
| ALE-Bench | 996.5 | — |
| AlgoTune | 1.47 | — |
| HumanEval+ | — | 86.6% |
| MBPP+ | — | 73% |

## Agentic & Tool Use

- Claude Opus 4.6: 51.0 (#4)
- DeepSeek-V3: —

| Benchmark | Claude Opus 4.6 | DeepSeek-V3 |
|---|---|---|
| METR Time Horizons | 78.9% | 49.6% |
| Terminal-Bench | 79.8% | — |
| APEX-Agents | 46.3% | — |
| Remote Labor Index | 4.2% | — |
| τ²-bench Banking | 27.3% | — |
| Cybench | 93% | — |
| DeepResearch Bench | 55.3% | — |
| GBAEval | 44.1% | — |
| LMArena Search | 1253 | — |
| Vending-Bench 2 | 8,018 | — |

## Reasoning

- Claude Opus 4.6: 57.7 (#21)
- DeepSeek-V3: 20.0 (#226)

| Benchmark | Claude Opus 4.6 | DeepSeek-V3 |
|---|---|---|
| SimpleBench | 67.6% | 27.2% |
| Kagi LLM Benchmark | 83.6% | 52.3% |
| LMArena Hard Prompts | 1527 | 1365 |
| DTBench | 91.2% | 64.8% |
| LMCA | 55.8% | 15.5% |
| Epoch Capabilities Index | 155.36 | 135.95 |
| ForecastBench | 60 | 59.1 |
| ARC-AGI-2 | 69.2% | — |
| NYT Connections (extended) | 92.1% | — |
| ARC-AGI-1 | 94% | — |
| CritPt | — | 0% |
| Chess Puzzles | 17% | — |
| EnigmaEval | 7.6% | — |
| Thematic Generalization | 80.6% | — |
| EBR-Bench | 12.7% | — |
| LiveBench Reasoning | — | 65.8% |
| Mystery Game Puzzles | 25% | — |
| LiveBench Data Analysis | — | 60.9% |
| BIG-Bench Hard | — | 87.5% |
| HellaSwag | — | 88.9% |
| LiveBench | — | 66.9% |
| PIQA | — | 84.7% |
| WinoGrande | — | 85.2% |

## Math

- Claude Opus 4.6: 62.7 (#25)
- DeepSeek-V3: 30.4 (#208)

| Benchmark | Claude Opus 4.6 | DeepSeek-V3 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 94.4% | 37.8% |
| LMArena Math | 1517 | 1373 |
| FrontierMath (Feb 2025 set) | 40.7% | 1.7% |
| FrontierMath (Tiers 1-3) | 66% | — |
| FrontierMath Tier 4 | 26.8% | — |
| MathArena Final-Answer Competitions | 78.5% | — |
| ProofBench | 50% | — |
| Omni-MATH | — | 40.3% |
| LiveBench Math | — | 73.5% |
| MATH Level 5 | — | 75.5% |
| FrontierMath Tier 4 (v1) | 22.9% | — |

## Knowledge

- Claude Opus 4.6: 61.8 (#23)
- DeepSeek-V3: 37.5 (#147)

| Benchmark | Claude Opus 4.6 | DeepSeek-V3 |
|---|---|---|
| GPQA Diamond | 90.5% | 67.6% |
| Vectara Hallucination Rate | 12.2% | 6.1% |
| LMArena Expert | 1546 | 1352 |
| Humanity's Last Exam | 34.4% | — |
| SimpleQA Verified | 47% | — |
| MMLU-Pro | — | 72.3% |
| Confabulations | — | 26.1% |
| GPQA (HELM) | — | 53.8% |
| ARC (AI2) Challenge | — | 95.3% |
| MMLU | — | 87.2% |
| TriviaQA | — | 82.9% |

## Multimodal

- Claude Opus 4.6: 37.0 (#67)
- DeepSeek-V3: —

| Benchmark | Claude Opus 4.6 | DeepSeek-V3 |
|---|---|---|
| LMArena Vision | 1316 | — |
| Furniture Assembly | 28.3% | — |
| LMArena Document | 1507 | — |

## Multilingual

- Claude Opus 4.6: 58.2 (#4)
- DeepSeek-V3: 48.8 (#138)

| Benchmark | Claude Opus 4.6 | DeepSeek-V3 |
|---|---|---|
| LMArena Non-English | 1490 | 1358 |
| LMArena Chinese | 1551 | 1390 |
| LMArena French | 1513 | 1386 |
| LMArena German | 1504 | 1374 |
| LMArena Japanese | 1484 | 1333 |
| LMArena Korean | 1465 | 1321 |
| LMArena Russian | 1497 | 1372 |
| LMArena Spanish | 1512 | 1358 |

## Instruction Following

- Claude Opus 4.6: 79.7 (#3)
- DeepSeek-V3: 73.0 (#125)

| Benchmark | Claude Opus 4.6 | DeepSeek-V3 |
|---|---|---|
| LMArena Instruction Following | 1523 | 1345 |
| LiveBench Instruction Following | — | 81.5% |
| IFEval | — | 83.2% |

## Long Context

- Claude Opus 4.6: 48.1 (#13)
- DeepSeek-V3: 34.0 (#248)

| Benchmark | Claude Opus 4.6 | DeepSeek-V3 |
|---|---|---|
| LMArena Longer Query | 1520 | 1353 |
| Fiction.LiveBench | — | 50% |
| CL-bench | 20.7% | — |
| CL-bench Life | 17% | — |

## Writing & Preference

- Claude Opus 4.6: 73.7 (#9)
- DeepSeek-V3: 57.6 (#124)

| Benchmark | Claude Opus 4.6 | DeepSeek-V3 |
|---|---|---|
| LMArena Text | 1504 | 1375 |
| LMArena Creative Writing | 1506 | 1364 |
| EQ-Bench Creative Writing | 1809 | 1472 |
| LMArena Multi-Turn | 1514 | 1389 |
| Short-Story Creative Writing | — | 77% |
| WildBench | — | 83% |
| EQ-Bench 4 | 1223 | — |
| LiveBench Language | — | 49.1% |

## FAQ

### Is Claude Opus 4.6 better than DeepSeek-V3?

Claude Opus 4.6 is the stronger model overall, scoring 58.3 to 39.2 on the Noometry Index. DeepSeek-V3 costs 25× less per token, which makes it the better buy when Claude Opus 4.6's lead doesn't matter for your workload.

### Which is cheaper, Claude Opus 4.6 or DeepSeek-V3?

DeepSeek-V3 is cheaper. It lists at $0.24 per million input tokens and $0.90 per million output tokens; Claude Opus 4.6 lists at $5 and $25.

### Is Claude Opus 4.6 or DeepSeek-V3 better for coding?

Claude Opus 4.6 scores higher on coding benchmarks: 57.7 versus 41.9 in the Noometry coding category.

### Which has the bigger context window?

Claude Opus 4.6 does, with 1M tokens against 164K.

### How many benchmarks do Claude Opus 4.6 and DeepSeek-V3 share?

30 benchmarks have published results for both models. Claude Opus 4.6 has 68 scored results on Noometry and DeepSeek-V3 has 60.