What makes Gemini 4 Carbon different
Gemini 4 Carbon is built for the hardest problems, where checking every step matters more than a fast answer. Here is what stands out when you run it inside Doraverse.
Top-ranked reasoning
Google reported Gemini 4 Carbon at 59 on the Artificial Analysis Intelligence Index v5.0, ranked first at launch, ahead of Gemini 4 Argon at 50.
Stronger long-horizon engineering
Carbon reaches 65.2% on Terminal-Bench 4.0, up from Argon's 57.1%, and Google's teams have seen it keep working through dead ends for days on complex bugs.
State-of-the-art security work
It leads on SEC-bench Pro (88.2%) for finding vulnerabilities, ExploitGym (54.8%) for validating them, and CWE-bench (71.0%) for patching them.
Reasoning across the full context
On GraphWalks 256K to 1M, which tests reasoning across the whole context window, Carbon scores 96.2% against Argon's 84.2%.