Gemini 4 Carbon has taken first place in the MathArena benchmark

Google has released the Gemini 4 Carbon model, which has topped the MathArena math benchmark. The development surpassed the previous leader, GPT-6.1 Sol from OpenAI.

Anthropic's Fable 5.2 model also performed well in the testing, joining Google's product in displacing OpenAI from the top of the rankings. However, both new developments significantly exceed their competitor in terms of task execution costs.

The new model from Google will be significantly more expensive for users. According to test data, Gemini 4 Carbon costs **15 times more** than GPT-6.1 Sol.

During the trials, Google completely ignored the established time limit and almost never refused to perform complex tasks.

Such results show that leadership in specialized benchmarks is achieved at the cost of a multiple increase in computational expenses and processing time.