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. | 236B | 7B/13B/70B |
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 | 4k |
| Benchmarks | ||
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. | 87.2% | 29.9% |
SWE-Bench VerifiedCoding — Real coding tasks pulled from open-source projects — the AI has to find and fix actual bugs. A human-checked version of the original SWE-Bench. Higher is better. | 19% | — |
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. | 39.7% | — |
MATHReasoning — accuracy on the original competition mathematics test set | — | 13.5% |
ARC-ChallengeReasoning — multiple-choice science questions from the AI2 Reasoning Challenge; unrelated to ARC-AGI | — | 57.4% |
HellaSwagReasoning — commonsense completion of short scenarios | — | 85.3% |
WinoGrandeReasoning — resolving ambiguous pronouns in context | — | 80.2% |
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. | — | 68.9% |
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. | — | 56.8% |
| Overview | ||
| Company | DeepSeek | Meta |
| Release date | Jul 24 2024 | Jul 18 2023 |
| Access | Open Weight | Open Weight |
| Model details | View model | View model |
Other comparisons
DeepSeek Coder V2-0724vsClaude Fable 5.2LLaMA 2vsClaude Fable 5.2DeepSeek Coder V2-0724vsGPT-6.2 AstraLLaMA 2vsGPT-6.2 AstraDeepSeek Coder V2-0724vsGemini 4 HydrogenLLaMA 2vsGemini 4 HydrogenDeepSeek Coder V2-0724vsGrok 4.7LLaMA 2vsGrok 4.7DeepSeek Coder V2-0724vsMistral Large 4LLaMA 2vsMistral Large 4DeepSeek Coder V2-0724vsKimi K3.1 ProLLaMA 2vsKimi K3.1 ProFrequently asked questions
DeepSeek Coder V2-0724 leads LLaMA 2 on 1 of the 1 benchmark they both report (HumanEval). LLaMA 2 shipped 372 days before DeepSeek Coder V2-0724, so benchmark comparisons should account for the intervening progress.
DeepSeek Coder V2-0724 has 236B parameters, while LLaMA 2 has 7B/13B/70B. Context windows are 128k (DeepSeek Coder V2-0724) vs 4k (LLaMA 2).
On HumanEval, DeepSeek Coder V2-0724 leads at 87.2% vs LLaMA 2 at 29.9%.