Skip to content
TheChatArchiveThe Chat Archive
Claudeclaude-sonnet-4-20250514· AnthropicHugging Face LMArena Human Preference 140K

Which opensource models in leaderboard for maths

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Geminigemini-2.5-pro’s answer.
  1. User

    Which opensource models in leaderboard for maths

  2. Claude Assistant

    Several open-source models perform well on mathematical reasoning benchmarks. Here are some notable ones based on recent leaderboards:

    Strong performers on math benchmarks:

    Qwen2.5-Math series - Specifically designed for mathematical reasoning, these models (7B, 72B variants) show excellent performance on GSM8K, MATH, and other math benchmarks.

    DeepSeek-Math - The 7B model demonstrates strong mathematical problem-solving capabilities and often ranks highly on math-focused evaluations.

    MetaMath series - Built on Llama foundations but fine-tuned extensively for mathematical reasoning, showing good results on standard math benchmarks.

    Llama 3.1 models (8B, 70B, 405B) - The larger variants perform well on mathematical reasoning tasks, though the 405B model is quite resource-intensive.

    Mistral models - Particularly Mixtral 8x7B and newer Mistral variants show solid mathematical reasoning capabilities.

    Yi-Math - Another math-specialized model that performs competitively on mathematical benchmarks.

    The specific rankings can vary depending on which benchmark you're looking at (GSM8K, MATH, Hungarian high school finals, etc.) and when the evaluation was conducted. Math-specialized models like Qwen2.5-Math and DeepSeek-Math tend to outperform general-purpose models of similar size on mathematical tasks.

    Which specific mathematical tasks or benchmarks are you most interested in?

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.