MathArena: Gemini 4 Carbon Leads, 15 Times More Expensive than Sol

Google's model has taken the top spot in the MathArena benchmark, ahead of GPT-6.1 Sol and Claude Fable 5.2. This lead comes with frequent time overruns and a significantly higher cost with the reasoning parameters tested, while some of the OpenAI models mentioned are not yet publicly available.

Time Overruns and Rising Costs Under Tested Settings

The most common failure mode of Gemini 4 Carbon is exceeding the time limit. The model rarely abandons a task, which can be seen as either an advantage or a disadvantage. In one instance, a script ran for over eleven hours without any progress check. With the reasoning parameters used during testing, Gemini 4 Carbon costs about 15 times more than GPT-6.1 Sol, and approximately 40 times more than Claude Fable 5.2.

Limited Public Access for Some OpenAI Models

According to some reports, models developed by OpenAI have recently been able to solve complex equations such as the Navier-Stokes equations; however, these models are not yet available to the public. In this context, MathArena serves as a benchmark to measure the mathematical capabilities of accessible models.

Current Ranking and Announced Milestones of the Leaderboard

Gemini 4 Carbon, developed by Google, has become the first non-OpenAI model to top the MathArena benchmark in some time, pushing GPT-6.1 Sol out of first place. Claude Fable 5.2 holds the second position, and Gemini 4 Carbon is among the highest-performing public models for mathematical tasks. Expected performance milestones are set for November 10, 2026, for Gemini 4 Carbon (high), November 9, 2026, for Claude-Fable-5.2 (max), September 29, 2026, for GPT-6.1 Sol (max), and October 23, 2026, for Kimi K3.1 Pro (max).