Open-weight LLM comparison
DeepSeek R1 vs Llama 3.1 405B
Side-by-side
Straight from each model's release page and Hugging Face card. Memory shown is for the weights alone: add roughly 15% for real serving. Bold column marks the more parameters and the smaller memory footprint.
| Field | DeepSeek R1 | Llama 3.1 405B |
|---|---|---|
| Total parameters | 671 B | 405 B |
| Active parameters (MoE) | 37 B | — |
| Architecture | MoE (671B total, 37B active per token) | dense |
| Vendor | DeepSeek | Meta |
| License | MIT | Llama 3.1 Community License |
| Released | 2025-01-20 | 2024-07-23 |
| Weights @ FP16 | 1342 GB | 810 GB |
| Weights @ INT8 | 671 GB | 405 GB |
| Weights @ INT4 | 336 GB | 203 GB |
Which is smarter (published benchmarks)
Vendor-reported quality scores on the standard leaderboards. Bold column marks the higher score on the same test.
| Benchmark | DeepSeek R1 | Llama 3.1 405B |
|---|---|---|
| MMLU (5-shot) | 90.8 | 88.6 |
| MATH (AIME 2024, pass@1) | 79.8 | not published |
| GPQA (Diamond) | 71.5 | not published |
| Codeforces (Elo) | 2029.0 | not published |
| HumanEval | not published | 89.0 |
| MATH (0-shot) | not published | 73.8 |
| GPQA | not published | 51.1 |
Sources: DeepSeek R1 model card · Llama 3.1 405B model card. Benchmark methodology and prompt template can shift these numbers by several points, so treat these as relative rankings, not absolute scores.
Common questions
DeepSeek R1 vs Llama 3.1 405B: which is bigger?
DeepSeek R1 has more parameters (DeepSeek R1: 671 B; Llama 3.1 405B: 405 B). More parameters usually means higher ceiling on capability and higher memory requirement, though MoE architectures decouple total parameters from per-token compute.
DeepSeek R1 vs Llama 3.1 405B: which is newer?
DeepSeek R1 released 2025-01-20; Llama 3.1 405B released 2024-07-23.
DeepSeek R1 vs Llama 3.1 405B: which needs less memory to serve?
Llama 3.1 405B needs less HBM. Weights-only footprint at FP16: DeepSeek R1 1342 GB; Llama 3.1 405B 810 GB. Half those numbers at INT8, quarter at INT4. Real serving adds 10-30% for KV cache.
DeepSeek R1 vs Llama 3.1 405B: which license is more permissive?
DeepSeek R1: MIT. Llama 3.1 405B: Llama 3.1 Community License. Apache 2.0 and MIT allow unrestricted commercial use; Llama Community License allows commercial use but restricts training larger models on outputs; CC-BY-NC and vendor-specific licenses (Qwen 72B, Gemma) have narrower terms. Check the model card for the exact clauses.
See also: every LLM comparison · DeepSeek R1 full page · Llama 3.1 405B full page.