Compare AI models

Specifications
Parameters
2B/7B7B
Context window
8k32k
Benchmarks
HumanEval
32.3%38.4%
MATH
24.3%13.2%
Overview
CompanyGoogleMistral
Release dateFeb 21 2024May 22 2024
AccessOpen WeightOpen Weight
Model detailsView modelView model

Frequently asked questions

Gemma 1 and Mistral 7B v0.3 are evenly matched across the 2 benchmarks they both report (HumanEval, MATH). Gemma 1 shipped 91 days before Mistral 7B v0.3, so benchmark comparisons should account for the intervening progress.

Gemma 1 has 2B/7B parameters, while Mistral 7B v0.3 has 7B. Context windows are 8k (Gemma 1) vs 32k (Mistral 7B v0.3).

On HumanEval, Mistral 7B v0.3 leads at 38.4% vs Gemma 1 at 32.3%. On MATH, Gemma 1 leads at 24.3% vs Mistral 7B v0.3 at 13.2%.