Compare AI models

Specifications
Parameters
16B—
Context window
32k128k
Benchmarks
HumanEval
57.3%74.1%
MATH
27.9%50.6%
MMLU
55.7%81.3%
GSM8K
72%90%
Overview
CompanyDeepSeekSpaceXSI
Release dateMay 16 2024Mar 28 2024
AccessOpen WeightClosed
Model detailsView modelView model

Frequently asked questions

Grok‑1.5 leads DeepSeek V2 Lite on 4 of the 4 benchmarks they both report (HumanEval, MATH, MMLU, GSM8K). Grok‑1.5 shipped 49 days before DeepSeek V2 Lite, so benchmark comparisons should account for the intervening progress.

Context windows are 32k (DeepSeek V2 Lite) vs 128k (Grok‑1.5). DeepSeek V2 Lite is open weight, while Grok‑1.5 is closed.

On HumanEval, Grok‑1.5 leads at 74.1% vs DeepSeek V2 Lite at 57.3%. On MATH, Grok‑1.5 leads at 50.6% vs DeepSeek V2 Lite at 27.9%. On MMLU, Grok‑1.5 leads at 81.3% vs DeepSeek V2 Lite at 55.7%. On GSM8K, Grok‑1.5 leads at 90% vs DeepSeek V2 Lite at 72%.