Gemini 4 Carbon: benchmarks, price and rank
Gemini 4 Carbon from Google is 2nd of 123 rated AI models on Spezia's leaderboard, strongest on APEX-Agents and LMArena text. It costs $2 in, $10 out per million tokens.
- 2nd of 123 rated models
- Rating 2,266
- First on 5 of 12 benchmarks
- Closed
- Released Oct 28, 2026
- 1 million tokens of context
Where it is strongest
- 1st of 56 on APEX-Agents
- 1st of 107 on LMArena text
- 1st of 60 on Vending-Bench 2
- 2nd of 78 on SimpleQA Verified
- 2nd of 68 on SimpleBench
Where it falls behind
In the top half on every benchmark it has a result on.
Price: $2 in, $10 out per million tokens.
Its headline results
The benchmarks that count most in the rating, with its place among the models measured the same way, and where each number came from.
- APEX-Agents84.2%
Long tasks written by investment banking analysts, management consultants and corporate lawyers from their own work, done across files and apps in a realistic work setup.
1st of 56. Independent result from Epoch AI.
- FrontierSWE61.2%
Open-ended technical projects that take hours to tens of hours, from making systems faster to building large programs and applied machine learning research. Each run gets a 20-hour budget.
5th of 21. Independent result from Epoch AI.
- Vending-Bench 2$16,484
The model runs a simulated vending machine business for a year on its own: finding suppliers, ordering stock, setting prices and paying the daily fees.
1st of 60. Independent result from Epoch AI.
- CursorBench60.4%
Ambiguous coding tasks that span many files, taken from real sessions in the Cursor code editor.
1st of 20. Independent result from Epoch AI.
- SimpleBench92.6%
Multiple-choice puzzles about everyday situations, space, time and people, written to trip up models that pattern-match instead of think. Ordinary adults do well on them.
2nd of 68. Independent result from Epoch AI.
- SimpleQA Verified77.7%
Short factual questions with one checked answer, asked without web search.
2nd of 78. Independent result from Epoch AI.
- FrontierMath90.5%
Hundreds of new, unpublished math problems written by mathematicians, from hard undergraduate level up to research, with answers a computer can check.
6th of 76. Independent result from Epoch AI.
- LMArena text1,538
People put a question to two anonymous models and vote for the better answer.
1st of 107. Independent result from LMArena.
- LMArena WebDev1,768
People ask two anonymous models to build the same web page or app and vote for the better one.
4th of 98. Independent result from LMArena.
- ExploitGym53.5%
Real vulnerabilities in ordinary programs, Google's V8 JavaScript engine and the Linux kernel. Given the bug, the model has to write a working exploit.
1st of 8. The maker's own number, from Google, via LLM Stats.
4 more results in its full profile.
Every result with its source, what it costs per task, what jumped with its release, side by side with any other model, and answers to your questions about when to use it.
Models ranked near it
How the rating works, and every benchmark it uses: the model leaderboard.