‹ PublicAI IndexGoogle
The LLM benchmark aggregator.
Gemini 4 Carbon
Strongest in Human preference (#1 of 354), weakest in Computer use (#19 of 69). Above par in 17 of 17 scopes. Among the models it meets almost everywhere, it finishes behind Claude Opus 5 and Claude Opus 5.5 and ahead of Gemini 4 Argon and GPT-6 Astra.
§ 1 · Profile
What it is good at
Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.
Human preference68.8leads#1/3541/1
Human preference68.8leads#1/3541/1
Core abilities65.6leads#1/2002/4
Data analysis66.1leads#1/321/1
General intelligence68.4−0.6#2/1802/3
Agents63.3−3#7/2983/6
Web research60.7−2.9#4/361/2
Long-horizon work68.3−5.5#5/321/1
Knowledge work67.1−3.6#10/2082/2
Computer use57.7−10.4#19/691/2
Reasoning62.3−3.1#9/1953/5
Reasoning66.1−2.9#3/1473/3
Coding59.6−6.5#12/1792/6
Agentic coding60.3−6.4#10/1732/5
Professional56.8−5.6#17/1942/2
Document design60.8−7.6#5/321/1
Legal61.3−4.6#8/1731/1
§ 2 · Head to head
What it beats, and what beats it
The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for Gemini 4 Carbon, left for the other.
§ 3 · Sources
Where the numbers come from
8 publications, 17 figures. Every one links to the page it was read from.
LMArena Text 1526
Artificial Analysis Intelligence Index 60
GDPval-AA 1733
AA-Briefcase 1611
ARC-AGI-2 92.1%
LiveBench 83.5LiveBench · Agentic 72.7LiveBench · Reasoning 90.6LiveBench · Coding 91LiveBench · Browser Use 80LiveBench · Data Analysis 86.5LiveBench · Research 92.5LiveBench · Design 80.2
SimpleBench 92.6%
Vals · Legal Research Bench 55.77%Vals · Harvey Legal Agent Benchmark 17.5%Vals · Vibe Code Bench 89.06% (OpenHands)
Badge
[](https://publicai.io/model-index/m/gemini-4-carbon)