Model comparison
Claude Opus 4 vs DeepSeek V4 Pro
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 43.1 on the Noometry Index.
Last verified . 31 shared benchmarks.
DeepSeekSummary
- They share 31 benchmarks with published results for both. Claude Opus 4 scores higher in 2 categories and DeepSeek V4 Pro in 7 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where DeepSeek V4 Pro leads 56.5 to 27.3.
- The biggest single-benchmark swing is ARC-AGI-1: 35.7% for Claude Opus 4 and 90.5% for DeepSeek V4 Pro.
- DeepSeek V4 Pro is cheaper at $0.66 / $1.98 per million input/output tokens, against $15 / $75 for Claude Opus 4.
- DeepSeek V4 Pro accepts more context: 1M tokens versus 200K.
- DeepSeek V4 Pro has downloadable open weights; the other is API-only.
Side by side
| Claude Opus 4 | DeepSeek V4 Pro | |
|---|---|---|
| Provider | Anthropic | DeepSeek |
| Noometry Index | 43.1 | 54.3 |
| Released | 2025-05-22 | 2026-04-24 |
| Weights | Proprietary | Open |
| Context window | 200K | 1M |
| Max output | 32K | 393K |
| Input $ / M tokens | $15 | $0.66 |
| Output $ / M tokens | $75 | $1.98 |
| Results tracked | 56 | 48 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek V4 Pro leads
Claude Opus 4: 47.2 (#62), DeepSeek V4 Pro: 52.4 (#34)
| Benchmark | Claude Opus 4 | DeepSeek V4 Pro |
|---|---|---|
| SWE-bench Verified | 70.7% | 77.6% |
| WeirdML | 43.7% | 66.2% |
| LMArena Coding | 1442 | 1470 |
| FrontierCode | — | 28.6% |
| SWE-bench Verified (bash only) | 67.6% | — |
| Aider Polyglot | 72% | — |
| LMArena WebDev | — | 1582 |
| SciCode | — | 51% |
| GSO | 6.9% | — |
| ALE-Bench | — | 1,403 |
| AlgoTune | 1.33 | — |
Agentic & Tool Use Claude Opus 4 leads
Claude Opus 4: 34.8 (#42), DeepSeek V4 Pro: 32.8 (#58)
| Benchmark | Claude Opus 4 | DeepSeek V4 Pro |
|---|---|---|
| APEX-Agents | — | 47.3% |
| Cybench | 38% | — |
| DeepResearch Bench | 46.8% | — |
| LMArena Search | 1127 | — |
| METR Time Horizons | 63.9% | — |
| Vending-Bench 2 | — | 3,285 |
Reasoning DeepSeek V4 Pro leads
Claude Opus 4: 27.3 (#121), DeepSeek V4 Pro: 56.5 (#24)
| Benchmark | Claude Opus 4 | DeepSeek V4 Pro |
|---|---|---|
| ARC-AGI-2 | 8.6% | 61.3% |
| Kagi LLM Benchmark | 74.3% | 53.5% |
| ARC-AGI-1 | 35.7% | 90.5% |
| CritPt | 0.3% | 18% |
| LMArena Hard Prompts | 1399 | 1461 |
| DTBench | 81.6% | 93.9% |
| LMCA | 37.4% | 45.5% |
| Epoch Capabilities Index | 142.67 | 155.31 |
| ForecastBench | 61.1 | 56.1 |
| SimpleBench | 58.8% | — |
| NYT Connections (extended) | — | 91.3% |
| Chess Puzzles | — | 47% |
| EnigmaEval | 5.6% | — |
| Mystery Game Puzzles | — | 43% |
| Surface Evolver Bench | — | 40% |
Math DeepSeek V4 Pro leads
Claude Opus 4: 42.0 (#86), DeepSeek V4 Pro: 64.8 (#30)
| Benchmark | Claude Opus 4 | DeepSeek V4 Pro |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 64.4% | 98.6% |
| LMArena Math | 1390 | 1455 |
| FrontierMath (Tiers 1-3) | — | 64.6% |
| FrontierMath Tier 4 | — | 26.8% |
| MathArena Final-Answer Competitions | — | 76.6% |
| ProofBench | — | 50% |
| Omni-MATH | 61.6% | — |
| MATH Level 5 | 85% | — |
| FrontierMath (Feb 2025 set) | 4.5% | — |
| FrontierMath Tier 4 (v1) | 4.2% | — |
Knowledge DeepSeek V4 Pro leads
Claude Opus 4: 44.0 (#88), DeepSeek V4 Pro: 59.5 (#31)
| Benchmark | Claude Opus 4 | DeepSeek V4 Pro |
|---|---|---|
| GPQA Diamond | 76.3% | 91.7% |
| Vectara Hallucination Rate | 12% | 8.6% |
| LMArena Expert | 1386 | 1464 |
| Humanity's Last Exam | 10.7% | — |
| SimpleQA Verified | — | 52.9% |
| MMLU-Pro | 87.5% | — |
| Confabulations | 15.9% | — |
| GPQA (HELM) | 70.8% | — |
Multimodal Not comparable
Claude Opus 4: 31.5 (#106), DeepSeek V4 Pro: —
| Benchmark | Claude Opus 4 | DeepSeek V4 Pro |
|---|---|---|
| LMArena Vision | 1192 | — |
| GeoBench | 49% | — |
| VPCT | 38% | — |
Multilingual DeepSeek V4 Pro leads
Claude Opus 4: 48.8 (#138), DeepSeek V4 Pro: 54.4 (#45)
| Benchmark | Claude Opus 4 | DeepSeek V4 Pro |
|---|---|---|
| LMArena Non-English | 1362 | 1439 |
| LMArena Chinese | 1386 | 1486 |
| LMArena French | 1372 | 1472 |
| LMArena German | 1391 | 1458 |
| LMArena Japanese | 1331 | 1445 |
| LMArena Korean | 1321 | 1447 |
| LMArena Russian | 1392 | 1453 |
| LMArena Spanish | 1389 | 1458 |
Instruction Following Too close to call
Claude Opus 4: 77.1 (#28), DeepSeek V4 Pro: 76.1 (#47)
| Benchmark | Claude Opus 4 | DeepSeek V4 Pro |
|---|---|---|
| LMArena Instruction Following | 1406 | 1448 |
| IFEval | 91.8% | — |
Long Context DeepSeek V4 Pro leads
Claude Opus 4: 39.6 (#172), DeepSeek V4 Pro: 45.0 (#51)
| Benchmark | Claude Opus 4 | DeepSeek V4 Pro |
|---|---|---|
| LMArena Longer Query | 1422 | 1458 |
| Fiction.LiveBench | 61.1% | — |
| CL-bench Life | — | 13.5% |
Writing & Preference DeepSeek V4 Pro leads
Claude Opus 4: 61.2 (#89), DeepSeek V4 Pro: 65.5 (#46)
| Benchmark | Claude Opus 4 | DeepSeek V4 Pro |
|---|---|---|
| LMArena Text | 1377 | 1451 |
| LMArena Creative Writing | 1387 | 1446 |
| EQ-Bench Creative Writing | 1580 | 1553 |
| LMArena Multi-Turn | 1396 | 1467 |
| Short-Story Creative Writing | 83.6% | — |
| WildBench | 85.2% | — |
| EQ-Bench 4 | — | 1166 |
Frequently asked questions
Is Claude Opus 4 better than DeepSeek V4 Pro?
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 43.1 on the Noometry Index.
Which is cheaper, Claude Opus 4 or DeepSeek V4 Pro?
DeepSeek V4 Pro is cheaper. It lists at $0.66 per million input tokens and $1.98 per million output tokens; Claude Opus 4 lists at $15 and $75.
Is Claude Opus 4 or DeepSeek V4 Pro better for coding?
DeepSeek V4 Pro scores higher on coding benchmarks: 52.4 versus 47.2 in the Noometry coding category.
Which has the bigger context window?
DeepSeek V4 Pro does, with 1M tokens against 200K.
How many benchmarks do Claude Opus 4 and DeepSeek V4 Pro share?
31 benchmarks have published results for both models. Claude Opus 4 has 56 scored results on Noometry and DeepSeek V4 Pro has 48.