Sunday, 15 November 2026, story 4 of 7
Google's Gemini 4 Carbon tops MathArena at 15 times GPT-6.1 Sol's cost
Claude Fable 5.2 ranks second and GPT-6.1 Sol third. Carbon scores 85.4% against Sol's 80.6%, at $15.70 versus $1.07.

What happened
Google's Gemini 4 Carbon (high) leads the MathArena math benchmark, according to results Jasper Dekoninck posted on X, as reported by The Decoder. It scores 85.4% ±3.0% at a listed cost of $15.70 ±$1.33. Anthropic's Claude Fable 5.2 (max) is second at 82.8% ±5.3% and $40.20 ±$2.59. OpenAI's GPT-6.1 Sol (max) is third at 80.6% ±2.7% and $1.07 ±$0.018.
Carbon's most common failure is exceeding the time limit. In one case it ran a script for more than eleven hours without checking progress. Moonshot AI's Kimi K3.1 Pro (max) is the best open model at 54.3%.
What it means
For math-heavy work, the table says accuracy has a steep price. Carbon's lead over Sol is about five points, while its listed cost is roughly 15 times higher. Fable 5.2 costs about 40 times as much for a score between the two. The margins also overlap, so the ordering of the top three is likely less firm than the ranks suggest.
Sol looks like the cost-efficient reference point. Carbon's habit of rarely giving up, and its time-limit failures, mean teams running it in agent loops may need hard timeouts and progress checks. The source does not give per-token prices or say how these results were run beyond the reasoning settings. It also notes that OpenAI's models, which reportedly solved harder equations, are not publicly available, so they are not part of this comparison.
What to do: Run Carbon and GPT-6.1 Sol on a sample of your own math tasks, with time limits and a cost cap, before paying for the top-ranked model.
Source
Filed under Research. Published .
"What it means" is analysis, not reporting. Found an error? Write to daily@vibitely.com.