← All news·2026-11-14·3 min read

Gemini 4 Carbon Tops MathArena, but Costs 15 Times More

Gemini 4 Carbon beats its rivals on MathArena, but running the benchmark costs nearly 15 times more than it does for GPT-6.1 Sol.

aigooglebenchmarks

Google released a new model called Gemini 4 Carbon on November 10. It topped MathArena, a benchmark that measures how well AI models solve math problems, scoring 85.4%. That beat Fable 5.2, a model from the AI company Anthropic, which scored 82.8%, and GPT-6.1 Sol, a model from OpenAI, which scored 80.6%. But running the benchmark on Carbon cost $15.70, while running it on Sol cost $1.07. That is almost fifteen times more.

Topping a benchmark no longer guarantees the best price. Companies that pick a model for heavy math or engineering work increasingly face a trade-off. They must choose between accuracy and the bill for every run. OpenAI's price for GPT-6.1 Sol suggests the company chose low cost over the top spot on the leaderboard.

Carbon's most common failure is not a wrong calculation. It is running past the time limit. In one case, the model ran a script for more than eleven hours without checking its own progress. How much that risk matters outside a benchmark test has not been checked yet.

Source: the-decoder.com

Free course

Stop reading about AI — start building with it

The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.

Start free →
EAEvgenii Arsentev

Author

Evgenii Arsentev

PhD · Founder & CEO, Dad4Kids