Gemini 4 Carbon tops MathArena but costs 15 times more than GPT-6.1 Sol

Biểu tượng Gemini 4 Carbon của Google trên bảng xếp hạng toán MathArena

Gemini 4 Carbon has overtaken GPT-6.1 Sol to take first place on the MathArena math leaderboard, but running the benchmark cost roughly 15 times more than its rival.

⚡ MIGOVI TL;DR

  • Google's Gemini 4 Carbon sits at the top of MathArena, knocking GPT-6.1 Sol out of first place.
  • Claude Fable 5.2 ranks second, placing it among the strongest publicly available models for math.
  • Carbon's most common failure is exceeding the time limit, and the model almost never gives up.
  • With the reasoning settings used in testing, Carbon and Fable 5.2 cost roughly 15 times and 40 times as much as GPT-6.1 Sol, respectively.

Gemini 4 Carbon climbs to the top of MathArena

Google's Gemini 4 Carbon is the first non-OpenAI model to claim the top spot on the MathArena math leaderboard in a long while. The result pushes GPT-6.1 Sol out of the lead.

Claude Fable 5.2 follows closely in second place. With these rankings, both can be considered among the strongest models for math tasks among those publicly released. OpenAI's models, reportedly able to crack tough problems such as the Navier-Stokes equations, are still not available to outside users.

How Carbon works, and where it falls short

The most common error for Gemini 4 Carbon is exceeding the time limit. The model also almost never quits midway. Depending on your perspective, that trait can be seen as either a strength or a weakness.

One case illustrates this clearly: Carbon launched a script that ran for more than 11 hours without ever checking its progress. This kind of persistence helps the model pursue a problem to the end, but it can also burn through time and resources.

Significantly higher costs

The stronger performance comes with a hefty price tag. With the reasoning configuration used in testing, Carbon costs roughly 15 times as much as GPT-6.1 Sol.

Claude Fable 5.2 is even pricier, at roughly 40 times the cost of GPT-6.1 Sol. That means Carbon's lead is not the most economical choice when judged on cost.

MIGOVI VIEWPOINT

A high MathArena ranking shows that publicly available models are advancing quickly in math. Still, the cost gap deserves careful thought. For compute-heavy tasks, users should set clear runtime limits and weigh accuracy against budget before choosing a model. The leaderboard also doesn't include OpenAI models that haven't been released, so the top spot could still change.

Source: The Decoder (AI)

Reviewed by: Hero Association

Share your thoughts ^^

Your email address will not be published. Required fields are marked *

Insights

Contents

  1. Gemini 4 Carbon climbs to the top of MathArena
  2. How Carbon works, and where it falls short
  3. Significantly higher costs