Compare AI models

Specifications
Parameters
671B70B
Context window
128k128k
API pricing
Cheapest input
—$0.10DeepInfra
Cheapest output
—$0.32DeepInfra
—$36.00DeepInfra · turbo · fp8
Benchmarks
GPQA Diamond
71.5%50.5%
MMLU-Pro
84%68.9%
MMLU
90.8%86%
Overview
CompanyDeepSeekMeta
Release dateJan 20 2025Dec 6 2024
AccessOpen WeightOpen Weight
Model detailsView modelView model

Frequently asked questions

DeepSeek R1 leads LLaMA 3.3 on 3 of the 3 benchmarks they both report (GPQA Diamond, MMLU-Pro, MMLU). LLaMA 3.3 shipped 45 days before DeepSeek R1, so benchmark comparisons should account for the intervening progress.

DeepSeek R1 has 671B parameters, while LLaMA 3.3 has 70B. Context windows are 128k (DeepSeek R1) vs 128k (LLaMA 3.3).

On GPQA Diamond, DeepSeek R1 leads at 71.5% vs LLaMA 3.3 at 50.5%. On MMLU-Pro, DeepSeek R1 leads at 84% vs LLaMA 3.3 at 68.9%. On MMLU, DeepSeek R1 leads at 90.8% vs LLaMA 3.3 at 86%.