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

Google's Gemini 4 Carbon has become the first non-OpenAI model in a while to top the MathArena math benchmark, knocking GPT-6.1 Sol from the top spot. Claude Fable 5.2 came in second. The result makes Gemini 4 Carbon potentially one of the most capable publicly available models for math tasks.

Benchmark result and availability

OpenAI's models, which recently allegedly solved demanding equations such as the Navier-Stokes equations, are not publicly available yet. Gemini 4 Carbon's most common failure mode is exceeding the time limit.

Persistence during tasks

The model also almost never gives up on a task, a trait that can be read as either positive or negative. In one case, it launched a script that ran for more than eleven hours without checking on its progress.

Cost comparison

With the reasoning settings used in testing, Carbon and Fable 5.2 cost roughly 15 and 40 times as much as GPT-6.1 Sol, respectively. Gemini 4 Carbon (high) had an expected cost of $15.70, plus or minus $1.33, while GPT-6.1 Sol (max) had an expected cost of $1.07, plus or minus $0.018. Claude-Fable-5.2 (max) had an expected cost of $40.20, plus or minus $2.59.

Rankings and release dates

The benchmark listing names Gemini 4 Carbon (high) as number one overall, with an expected performance of 85.4 percent, plus or minus 3.0 percent, and a Google release date of November 10, 2026. Claude-Fable-5.2 (max) is number two overall, with 82.8 percent, plus or minus 5.3 percent, and an Anthropic release date of November 9, 2026. GPT-6.1 Sol (max) is number three overall, with 80.6 percent, plus or minus 2.7 percent, and an OpenAI release date of September 29, 2026. The best open model listed is Kimi K3.1 Pro (max), with 54.3 percent, plus or minus 5.8 percent, a Moonshot AI release date of October 23, 2026, and an expected cost of $3.70, plus or minus $0.27.

Sources: The Decoder