Google DeepMind's Gemini 4 Carbon (High) ranked first in Arena's Text Arena, second in Agent Arena, and fourth in Code Arena: WebDev, according to the platform's weekly update.
Google DeepMind · Arena leaderboard
Gemini 4 Carbon (High) takes No. 1 in text and No. 2 in agents
In the platform's weekly update, the model ranked first in Text Arena, second in Agent Arena, and fourth in Code Arena: WebDev, reshaping the Pareto frontier across all three.
1st
Text Arena
General text evaluation
2nd
Agent Arena
Tool use and multi-step actions
4th
Code Arena: WebDev
Web development coding tasks
Placement by arena, drawn to scale
Column height equals placement. Shorter column = higher rank (1 block = 1st place).
What "Pareto frontier" means here
A model that moves the frontier changes the best available trade-offs at each cost level.
Trade-off axis: performance versus efficiency.
Also noted: a model called Terra became more efficient, though no details were available.
Read as a positive signal
Top placement in general text evaluation
Strong results in agentic tasks that use tools and multi-step actions
Competitive showing in web development coding
Caveats to keep in mind
One weekly ranking reflects a single evaluation snapshot
It does not measure production reliability, safety, or cost
No independent benchmarks, pricing, or availability details are available yet
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...