Skip to content
Trending
Hitechreports
Tip
AI1 min read

Gemini 4 Carbon vs GPT-6.1 Sol: MathArena crown won, price tag staggering

Gemini 4 Carbon beats GPT-6.1 Sol on MathArena — but costs 15× as much per query. No public release date. No API. Just raw math, steep compute, and a leaderboard about to change its rules.

By The Prompt

The AI Desk · (2 hours ago)

Reported fromoutlet, credited1
MathArena leaderboard showing Gemini 4 Carbon at #1
MathArena leaderboard showing Gemini 4 Carbon at #1Photo via The Decoder

Gemini 4 Carbon just claimed first place on MathArena — the most widely cited public benchmark for mathematical reasoning, dethroning GPT-6.1 Sol. It’s the first non-OpenAI model to top the leaderboard in months. Claude Fable 5.2 landed second. Neither OpenAI’s latest models nor their reported Navier-Stokes breakthroughs are publicly accessible yet, so they don’t appear on the live leaderboard at all.

MathArena scores measure correctness and reasoning depth on problems ranging from Olympiad-level algebra to formal theorem proving. Gemini 4 Carbon’s win is real. But it comes with a hard trade-off: cost. At the reasoning settings used in the test, Gemini 4 Carbon runs at roughly 15× the inference cost of GPT-6.1 Sol.

Fable 5.2 is even steeper, about 40× more expensive per query. That gap isn’t theoretical. It’s what developers would pay today, in real dollars, to run those models at scale. The Decoder notes Gemini 4 Carbon’s most frequent failure mode isn’t wrong answers, it’s timing out.

The model rarely abandons a problem. In one test case, it executed a script for over eleven hours without pausing to validate intermediate steps. That persistence may help on stubborn proofs, but it also inflates latency and cloud bills. No pricing or availability details were released by Google.

There’s no public API, no consumer-facing launch window, no hardware tie-in. This is a research-model benchmark result, not a product announcement. Same for Fable 5.2 and GPT-6.1 Sol: all three are currently accessible only via limited developer previews or academic sandboxes. What changes next?

MathArena’s methodology is under review after this result. The team confirmed it’s adding time-weighted scoring in early 2027 to penalize runaway inference. Until then, Gemini 4 Carbon holds the crown, and the bill.

TopicsGPT-6.1 Solfable-5-2ai-benchmarksmatharenaGemini 4 Carbon

Reported from

How this story was made

Written by the Hitechreports desk from the reporting credited above, with facts attributed to their original publishers. We do not test devices ourselves; anything about performance, battery life or cameras comes from the outlets that did. Prices are as reported at the time of writing. Editorial policy · Report an error

What did you make of it?
ShareWhatsAppXFacebookTelegram

Filed by The AI Desk

Models, assistants and the companies and chips behind them, reported from what was released and what was claimed, with the difference kept clear.

More from this desk →

Be the first to comment

Join the argument. No password, just your email or a passkey.

    Read next

    More AI →