Compare AI models
| Specifications | ||
ParametersThe learned values that shape how a model responds. B means billion and T means trillion. For mixture-of-experts models, the total count is larger than the number active for each token. Parameter count alone does not measure quality. | 16B/236B | 51B |
Context windowHow much information the model can work with in one request, measured in tokens — pieces of words or other data. This includes your question, attachments, conversation and reply. Limits can vary by model variant and provider. | 128k | 128k |
| Benchmarks | ||
ARC-ChallengeReasoning — multiple-choice science questions from the AI2 Reasoning Challenge; unrelated to ARC-AGI | 92.8% | 69.2% |
MMLUGeneral knowledge — A 57-subject multiple-choice exam — history, law, medicine, maths — that was the standard measure of how much a model knows from 2020 until roughly 2024, when frontier scores crowded into the high 80s and labs moved on to harder tests. The scores here were published years apart under different testing setups, so read them as a historical record rather than a like-for-like ranking. Higher is better. | 79.2% | 80.2% |
GSM8KGrade-school math — Grade-school maths word problems that take a few steps of arithmetic to work through. It separated the models of 2022 and 2023 sharply, then saturated. One caveat on the historical numbers: OpenAI included part of the GSM8K training set in GPT-4's pre-training mix, so GPT-4's score is not a clean few-shot result. Higher is better. | 94.9% | 91.43% |
LiveCodeBenchCompetitive coding — Coding problems published so recently the AI can't have seen them in training — a contamination-free test of raw programming skill. Higher is better. | 43.4% | — |
HumanEvalFunction synthesis — 164 small Python problems: the AI is given a function's description and has to write the working function. This was the coding benchmark of the GPT-3.5 and GPT-4 era, before the field moved to fixing real bugs in real repositories. Higher is better. | 90.2% | — |
MATHReasoning — accuracy on the original competition mathematics test set | 75.7% | — |
HellaSwagReasoning — commonsense completion of short scenarios | — | 85.58% |
WinoGrandeReasoning — resolving ambiguous pronouns in context | — | 84.53% |
| Overview | ||
| Company | DeepSeek | NVIDIA |
| Release date | Jun 14 2024 | Sep 23 2024 |
| Access | Open Weight | Open Weight |
| Model details | View model | View model |
Other comparisons
DeepSeek Coder V2vsClaude Fable 5.2Llama 3.1 Nemotron 51BvsClaude Fable 5.2DeepSeek Coder V2vsGPT-6.2 AstraLlama 3.1 Nemotron 51BvsGPT-6.2 AstraDeepSeek Coder V2vsGemini 4 NeonLlama 3.1 Nemotron 51BvsGemini 4 NeonDeepSeek Coder V2vsMuse EmberLlama 3.1 Nemotron 51BvsMuse EmberDeepSeek Coder V2vsGrok 4.7Llama 3.1 Nemotron 51BvsGrok 4.7DeepSeek Coder V2vsMistral Large 4Llama 3.1 Nemotron 51BvsMistral Large 4Frequently asked questions
DeepSeek Coder V2 leads Llama 3.1 Nemotron 51B on 2 of the 3 benchmarks they both report (ARC-Challenge, MMLU, GSM8K). DeepSeek Coder V2 shipped 101 days before Llama 3.1 Nemotron 51B, so benchmark comparisons should account for the intervening progress.
DeepSeek Coder V2 has 16B/236B parameters, while Llama 3.1 Nemotron 51B has 51B. Context windows are 128k (DeepSeek Coder V2) vs 128k (Llama 3.1 Nemotron 51B).
On ARC-Challenge, DeepSeek Coder V2 leads at 92.8% vs Llama 3.1 Nemotron 51B at 69.2%. On MMLU, Llama 3.1 Nemotron 51B leads at 80.2% vs DeepSeek Coder V2 at 79.2%. On GSM8K, DeepSeek Coder V2 leads at 94.9% vs Llama 3.1 Nemotron 51B at 91.43%.