Llama 3.1 Instruct 405B is the stronger all-rounder.
Scores updated · 61 tests both models report · How we compare
Where each one wins
Tests won in each of the two areas where both have results. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
Reasoning and long documents rest on a single test each.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where Llama 3.1 Instruct 405B pulls ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+16.2points ahead
- Questions about very long textsLongBench v2+4.5points ahead
Where DeepSeek-V2 pulls ahead
No clear win on a test scored out of 100.
Every test, side by side
All 61 tests both models report. The winning score is in its model's colour; marks a score checked independently.
ReasoningLlama 3.1 Instruct 405B
- GPQA DiamondLlama 3.1 Instruct 405B by 16.235.351.5+16.2
Long documentsLlama 3.1 Instruct 405B
- LongBench v2Llama 3.1 Instruct 405B by 4.531.636.1+4.5
Other results59 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- HumanEvalLlama 3.1 Instruct 405B by 45.743.389+45.7
- IFEvalLlama 3.1 Instruct 405B by 30.957.788.6+30.9
- MATHLlama 3.1 Instruct 405B by 30.443.473.8+30.4
- MGSMLlama 3.1 Instruct 405B by 2863.691.6+28
- MMLU-ProLlama 3.1 Instruct 405B by 21.951.473.3+21.9
- AIME 2024Llama 3.1 Instruct 405B by 18.74.623.3+18.7
- GSM8KLlama 3.1 Instruct 405B by 17.679.296.8+17.6
- MATH-500 (EM)Llama 3.1 Instruct 405B by 17.556.373.8+17.5
- LiveCodeBenchLlama 3.1 Instruct 405B by 16.111.627.7+16.1
- CCPMDeepSeek-V2 by 14.49378.6+14.4
- MMLULlama 3.1 Instruct 405B by 1375.688.6+13
- CMMLUDeepSeek-V2 by 10.38473.7+10.3
- CMMLU (Acc.)DeepSeek-V2 by 10.38473.7+10.3
- CRUXEval-O (output prediction)Llama 3.1 Instruct 405B by 10.149.859.9+10.1
- MMMLULlama 3.1 Instruct 405B by 9.86473.8+9.8
- MMMLU-non-English (Acc.)Llama 3.1 Instruct 405B by 9.86473.8+9.8
- C-EvalDeepSeek-V2 by 8.981.472.5+8.9
- MMLU-ReduxLlama 3.1 Instruct 405B by 8.377.986.2+8.3
- HumanEval-Mul (Pass@1)Llama 3.1 Instruct 405B by 7.969.377.2+7.9
- CodeforcesLlama 3.1 Instruct 405B by 7.817.525.3+7.8
- MGSM (EM)Llama 3.1 Instruct 405B by 6.363.669.9+6.3
- CRUXEval-I (input prediction)Llama 3.1 Instruct 405B by 652.558.5+6
- MMLU (Acc.)Llama 3.1 Instruct 405B by 678.484.4+6
- DROP (3-shot F1)Llama 3.1 Instruct 405B by 5.78388.7+5.7
- MMLU-Redux (Acc.)Llama 3.1 Instruct 405B by 5.775.681.3+5.7
- DROP (F1)Llama 3.1 Instruct 405B by 5.680.486+5.6
- MATH (EM)Llama 3.1 Instruct 405B by 5.643.449+5.6
- ARC-ChallengeLlama 3.1 Instruct 405B by 4.792.296.9+4.7
- DROPLlama 3.1 Instruct 405B by 4.780.184.8+4.7
- RACE-HighLlama 3.1 Instruct 405B by 4.252.656.8+4.2
- BBHLlama 3.1 Instruct 405B by 4.178.882.9+4.1
- BBH (EM)Llama 3.1 Instruct 405B by 4.178.882.9+4.1
- LiveCodeBench-Base (Pass@1)Llama 3.1 Instruct 405B by 3.911.615.5+3.9
- Aider-Edit (Acc.)Llama 3.1 Instruct 405B by 3.660.363.9+3.6
- MBPPLlama 3.1 Instruct 405B by 3.46568.4+3.4
- AGIEvalLlama 3.1 Instruct 405B by 3.157.560.6+3.1
- ARC-Challenge (Acc.)Llama 3.1 Instruct 405B by 3.192.295.3+3.1
- FRAMES (Acc.)Llama 3.1 Instruct 405B by 3.166.970+3.1
- NaturalQuestionsLlama 3.1 Instruct 405B by 2.938.641.5+2.9
- NaturalQuestions (EM)Llama 3.1 Instruct 405B by 2.938.641.5+2.9
- TriviaQA (EM)Llama 3.1 Instruct 405B by 2.78082.7+2.7
- C3Llama 3.1 Instruct 405B by 2.377.479.7+2.3
- C3 (Acc.)Llama 3.1 Instruct 405B by 2.377.479.7+2.3
- HellaSwagLlama 3.1 Instruct 405B by 2.187.189.2+2.1
- HellaSwag (Acc.)Llama 3.1 Instruct 405B by 2.187.189.2+2.1
- PIQALlama 3.1 Instruct 405B by 283.985.9+2
- PIQA (Acc.)Llama 3.1 Instruct 405B by 283.985.9+2
- Chinese SimpleQA (C-SimpleQA)Llama 3.1 Instruct 405B by 1.948.550.4+1.9
- GSM8K (EM)Llama 3.1 Instruct 405B by 1.981.683.5+1.9
- CMath (EM)DeepSeek-V2 by 1.478.777.3+1.4
- CMRCDeepSeek-V2 by 1.477.476+1.4
- MMLU-Pro (Acc.)Llama 3.1 Instruct 405B by 1.451.452.8+1.4
- RACE-MiddleLlama 3.1 Instruct 405B by 1.173.174.2+1.1
- WinoGrandeDeepSeek-V2 by 1.186.385.2+1.1
- WinoGrande (Acc.)DeepSeek-V2 by 1.186.385.2+1.1
- CLUEWSCLlama 3.1 Instruct 405B by 18283+1
- ARC-Easy (Acc.)tie97.698.4tie
- Pile-test (BPB)lower is bettertie0.60.5tie
- The Pile (Test, BPB)lower is bettertie0.60.5tie
Questions people ask
Which is better, DeepSeek-V2 or Llama 3.1 Instruct 405B?
Llama 3.1 Instruct 405B wins all two areas where both have results: reasoning and long documents. DeepSeek-V2 wins none.
How do you compare the two?
We use the 61 benchmark tests both models have published scores on. The verdict counts the 2 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 59 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.