Back to models

DeepSeek-V2 (MoE-236B, May 2024)

ECI#182
121
DeepSeek
May 7, 2024
Open weights

DeepSeek-V2 (MoE-236B, May 2024) is an open-weights model from DeepSeek, released May 7, 2024. It has an ECI score of 121 and ranks 182nd of 240 tracked models. At the time of its release, it ranked 6th of 54 tracked models.

DeepSeek-V2 (MoE-236B, May 2024) vs select alternatives

DomainBenchmark
DeepSeek-V2 (MoE-236B, May 2024)
DeepSeek
May 7, 2024
Open weights
DeepSeek Coder 33B
DeepSeek
Nov. 2, 2023
Open weights
phi-3-small 7.4B
Microsoft
Apr. 23, 2024
Open weights
Llama 3-8B
Meta AI
Apr. 18, 2024
Open weights
ECI scoreBest score 173
12177116109
Ranking#182#230#192#203
Conceptual reasoning
BBHBest score 89%
79%
n/a
79%
n/a
Language understanding
WinoGrandeBest score 89%
86%
62%
82%
76%
HellaSwagBest score 95%
87%
n/a
77%
n/a
Science
ARC AI2Best score 95%
92%
42%
91%
83%
World knowledge
TriviaQABest score 88%
80%
n/a
58%
68%
MMLUBest score 88%
78%
39%
76%
69%

Model metadata

General
OrganizationDeepSeek
Release dateMay 7, 2024
Model accessibilityOpen weights
CountryChina
Knowledge cutoffUnknown
Provider notesDeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model
Pricing
Input price—
Output price—
Training
Parameters2.4e11
Training compute1.0e24 FLOP
Training hardwareNVIDIA H800 SXM5
Training timeUnknown
Training dataset size8.1e12 datapoints