LLM Reference

WizardMath Models by WizardLM Team

WizardLM TeamLlama 2 CommunityOpen weightsMath
4 models2024

Last refreshed 2026-06-29. Next refresh: weekly.

Details

ResearcherWizardLM Team
Commercial useCommercial use: conditional
Models4
Released2024

About

WizardMath is a family of open-source large language models (LLMs) specifically designed for mathematical reasoning tasks. Developed by the WizardLM team, these models are built upon the LLaMA-2 architecture and fine-tuned using a novel technique called Reinforced Evol-Instruct.The WizardMath family includes models of various sizes, such as 7B, 13B, and 70B parameters. These models have demonstrated impressive performance on mathematical reasoning benchmarks like GSM8k and MATH, outperforming many other open-source LLMs and even some commercial models.

Decision facts

Best fit
mathmath-heavy prompts
Capability starting point
WizardMath 70B
Lowest tracked input
Not tracked
Closest related family
WizardCoder

Current Variants

Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.

4 in view

Use when the workload needs 70B parameters.

2024-0170B parameters

Use when the workload needs 13B parameters.

2024-0113B parameters

Use when the workload needs 7B parameters.

2024-017B parameters

Use when the workload needs 7B parameters.

2024-017B parameters

Release Timeline

1 release group
2024-01
4 current
WizardMath 13B
13B parameters
Current
WizardMath 70B
70B parameters
Current
WizardMath 7B
7B parameters
Current
WizardMath 7B V1.1
7B parameters
Current

Specifications(4 models)

WizardMath model specifications comparison
ModelReleasedParameters
WizardMath 70B2024-0170B
WizardMath 13B2024-0113B
WizardMath 7B2024-017B
WizardMath 7B V1.12024-017B