Compare AI models

Specifications
Parameters
—16B
Context window
200k32k
Benchmarks
HumanEval
75.9%57.3%
MMLU
76.7%55.7%
GSM8K
88.9%72%
Overview
CompanyAnthropicDeepSeek
Release dateMar 4 2024May 16 2024
AccessClosedOpen Weight
Model detailsView modelView model

Frequently asked questions

Claude 3 Haiku leads DeepSeek V2 Lite on 3 of the 3 benchmarks they both report (HumanEval, MMLU, GSM8K). Claude 3 Haiku shipped 73 days before DeepSeek V2 Lite, so benchmark comparisons should account for the intervening progress.

Context windows are 200k (Claude 3 Haiku) vs 32k (DeepSeek V2 Lite). Claude 3 Haiku is closed, while DeepSeek V2 Lite is open weight.

On HumanEval, Claude 3 Haiku leads at 75.9% vs DeepSeek V2 Lite at 57.3%. On MMLU, Claude 3 Haiku leads at 76.7% vs DeepSeek V2 Lite at 55.7%. On GSM8K, Claude 3 Haiku leads at 88.9% vs DeepSeek V2 Lite at 72%.