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. | 1.5B/7B/8B/14B/32B/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 |
| Benchmarks | ||
MATH-500Reasoning — accuracy on the 500-problem MATH subset | 94.5% | 87.1% |
AIME 2024Reasoning — single-answer accuracy on the 2024 invitational mathematics exam without tools | 70% | 36.7% |
GPQA DiamondScience — Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better. | 65.2% | 69.5% |
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. | — | 38% |
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. | 57.5% | — |
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. | — | 88.6% |
DROPReasoning — three-shot token F1 on questions requiring reasoning over paragraphs | — | 83.4 |
SimpleQAKnowledge — accuracy on short, fact-seeking questions in the original SimpleQA set | — | 62.5% |
MMMLUKnowledge — average accuracy on multilingual translations of MMLU | — | 85.1% |
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. | — | 90.8% |
MGSMMultilingual math — solving grade-school word problems translated into multiple languages | — | 86.9% |
MMMUMultimodal — Tests the AI on understanding images and text together across many college subjects. Higher is better. | — | 74.4% |
| Overview | ||
| Company | DeepSeek | OpenAI |
| Release date | Jan 20 2025 | Feb 27 2025 |
| Access | Open Weight | Closed |
| Model details | View model | View model |
Other comparisons
DeepSeek R1 DistillvsClaude Fable 5.2GPT-4.5vsClaude Fable 5.2DeepSeek R1 DistillvsGemini 4 HydrogenGPT-4.5vsGemini 4 HydrogenDeepSeek R1 DistillvsMuse EmberGPT-4.5vsMuse EmberDeepSeek R1 DistillvsGrok 4.7GPT-4.5vsGrok 4.7DeepSeek R1 DistillvsMistral Large 4GPT-4.5vsMistral Large 4DeepSeek R1 DistillvsKimi K3.1 ProGPT-4.5vsKimi K3.1 ProFrequently asked questions
DeepSeek R1 Distill leads GPT-4.5 on 2 of the 3 benchmarks they both report (MATH-500, AIME 2024, GPQA Diamond). DeepSeek R1 Distill shipped 38 days before GPT-4.5, so benchmark comparisons should account for the intervening progress.
DeepSeek R1 Distill is open weight, while GPT-4.5 is closed.
On MATH-500, DeepSeek R1 Distill leads at 94.5% vs GPT-4.5 at 87.1%. On AIME 2024, DeepSeek R1 Distill leads at 70% vs GPT-4.5 at 36.7%. On GPQA Diamond, GPT-4.5 leads at 69.5% vs DeepSeek R1 Distill at 65.2%.