Model comparison
DeepSeek V4 Pro vs Llama 3.1-405B
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 30.7 on the Noometry Index.
Last verified . 25 shared benchmarks.
DeepSeek
MetaSummary
- They share 25 benchmarks with published results for both. DeepSeek V4 Pro scores higher in 9 categories and Llama 3.1-405B in 0 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in math, where DeepSeek V4 Pro leads 64.8 to 18.4.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 98.6% for DeepSeek V4 Pro and 9.7% for Llama 3.1-405B.
Side by side
| DeepSeek V4 Pro | Llama 3.1-405B | |
|---|---|---|
| Provider | DeepSeek | Meta |
| Noometry Index | 54.3 | 30.7 |
| Released | 2026-04-24 | 2024-07-23 |
| Weights | Open | Open |
| Context window | 1M | — |
| Max output | 393K | — |
| Input $ / M tokens | $0.66 | — |
| Output $ / M tokens | $1.98 | — |
| Results tracked | 48 | 42 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek V4 Pro leads
DeepSeek V4 Pro: 52.4 (#34), Llama 3.1-405B: 33.1 (#262)
| Benchmark | DeepSeek V4 Pro | Llama 3.1-405B |
|---|---|---|
| WeirdML | 66.2% | 21.4% |
| LMArena Coding | 1470 | 1291 |
| SWE-bench Verified | 77.6% | — |
| FrontierCode | 28.6% | — |
| LMArena WebDev | 1582 | — |
| SciCode | 51% | — |
| ALE-Bench | 1,403 | — |
Agentic & Tool Use DeepSeek V4 Pro leads
DeepSeek V4 Pro: 32.8 (#58), Llama 3.1-405B: 21.0 (#140)
| Benchmark | DeepSeek V4 Pro | Llama 3.1-405B |
|---|---|---|
| APEX-Agents | 47.3% | — |
| TheAgentCompany | — | 7.4% |
| Cybench | — | 7.5% |
| Vending-Bench 2 | 3,285 | — |
Reasoning DeepSeek V4 Pro leads
DeepSeek V4 Pro: 56.5 (#24), Llama 3.1-405B: 16.8 (#300)
| Benchmark | DeepSeek V4 Pro | Llama 3.1-405B |
|---|---|---|
| Kagi LLM Benchmark | 53.5% | 45% |
| LMArena Hard Prompts | 1461 | 1269 |
| DTBench | 93.9% | 61.4% |
| Epoch Capabilities Index | 155.31 | 128.75 |
| ForecastBench | 56.1 | 59.9 |
| ARC-AGI-2 | 61.3% | — |
| SimpleBench | — | 23% |
| NYT Connections (extended) | 91.3% | — |
| ARC-AGI-1 | 90.5% | — |
| CritPt | 18% | — |
| Chess Puzzles | 47% | — |
| Mystery Game Puzzles | 43% | — |
| LMCA | 45.5% | — |
| Surface Evolver Bench | 40% | — |
| BIG-Bench Hard | — | 82.9% |
| HellaSwag | — | 89.2% |
| PIQA | — | 85.9% |
| WinoGrande | — | 89.2% |
Math DeepSeek V4 Pro leads
DeepSeek V4 Pro: 64.8 (#30), Llama 3.1-405B: 18.4 (#290)
| Benchmark | DeepSeek V4 Pro | Llama 3.1-405B |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 98.6% | 9.7% |
| LMArena Math | 1455 | 1281 |
| FrontierMath (Tiers 1-3) | 64.6% | — |
| FrontierMath Tier 4 | 26.8% | — |
| MathArena Final-Answer Competitions | 76.6% | — |
| ProofBench | 50% | — |
| Omni-MATH | — | 24.9% |
| MATH Level 5 | — | 49.8% |
Knowledge DeepSeek V4 Pro leads
DeepSeek V4 Pro: 59.5 (#31), Llama 3.1-405B: 30.4 (#227)
| Benchmark | DeepSeek V4 Pro | Llama 3.1-405B |
|---|---|---|
| GPQA Diamond | 91.7% | 50.9% |
| LMArena Expert | 1464 | 1243 |
| SimpleQA Verified | 52.9% | — |
| MMLU-Pro | — | 72.3% |
| Confabulations | — | 17.6% |
| Vectara Hallucination Rate | 8.6% | — |
| GPQA (HELM) | — | 52.2% |
| ARC (AI2) Challenge | — | 95.3% |
| MMLU | — | 84.5% |
| TriviaQA | — | 82.7% |
Multilingual DeepSeek V4 Pro leads
DeepSeek V4 Pro: 54.4 (#45), Llama 3.1-405B: 40.7 (#214)
| Benchmark | DeepSeek V4 Pro | Llama 3.1-405B |
|---|---|---|
| LMArena Non-English | 1439 | 1248 |
| LMArena Chinese | 1486 | 1242 |
| LMArena French | 1472 | 1279 |
| LMArena German | 1458 | 1252 |
| LMArena Japanese | 1445 | 1208 |
| LMArena Korean | 1447 | 1184 |
| LMArena Russian | 1453 | 1265 |
| LMArena Spanish | 1458 | 1260 |
Instruction Following DeepSeek V4 Pro leads
DeepSeek V4 Pro: 76.1 (#47), Llama 3.1-405B: 65.9 (#214)
| Benchmark | DeepSeek V4 Pro | Llama 3.1-405B |
|---|---|---|
| LMArena Instruction Following | 1448 | 1259 |
| IFEval | — | 81.1% |
Long Context DeepSeek V4 Pro leads
DeepSeek V4 Pro: 45.0 (#51), Llama 3.1-405B: 38.4 (#197)
| Benchmark | DeepSeek V4 Pro | Llama 3.1-405B |
|---|---|---|
| LMArena Longer Query | 1458 | 1266 |
| CL-bench Life | 13.5% | — |
Writing & Preference DeepSeek V4 Pro leads
DeepSeek V4 Pro: 65.5 (#46), Llama 3.1-405B: 38.9 (#251)
| Benchmark | DeepSeek V4 Pro | Llama 3.1-405B |
|---|---|---|
| LMArena Text | 1451 | 1284 |
| LMArena Creative Writing | 1446 | 1262 |
| EQ-Bench Creative Writing | 1553 | 870 |
| LMArena Multi-Turn | 1467 | 1297 |
| WildBench | — | 78.3% |
| EQ-Bench 4 | 1166 | — |
Frequently asked questions
Is DeepSeek V4 Pro better than Llama 3.1-405B?
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 30.7 on the Noometry Index.
Is DeepSeek V4 Pro or Llama 3.1-405B better for coding?
DeepSeek V4 Pro scores higher on coding benchmarks: 52.4 versus 33.1 in the Noometry coding category.
How many benchmarks do DeepSeek V4 Pro and Llama 3.1-405B share?
25 benchmarks have published results for both models. DeepSeek V4 Pro has 48 scored results on Noometry and Llama 3.1-405B has 42.