Not established from the available sources.
GOOGLE DEEPMIND / GOOGLE · MODEL RELEASE TRACKER
Carbon (Gemini 4 Carbon)
Gemini 4 Carbon is a frontier model announced by Google/DeepMind on 2026-10-28. Google describes it as their strongest Gemini 4 model with state-of-the-art cybersecurity capabilities and long-horizon (sustained agentic) workflows. Carbon is being rolled out initially to a subset of cyber defenders while undergoing pre-release testing, and it launches with the same introductory token pricing as Gemini 4 Argon.CURRENT SNAPSHOT2/5 DIMENSIONS WITH DATA
The dimensions that change the decision.
Not established from the available sources.
Not established from the available sources.
Artificial Analysis Intelligence Index (v5.0) rankingRanks #1 on Artificial Analysis Intelligence Index v5.0 with a score of 59.
Specific cybersecurity benchmark scoresReported scores include SEC-bench Pro 88.2% (finding vulnerabilities), ExploitGym 54.8% (exploitation), and CWE-bench 71.0% (patching).
Initial availability / rollout planGoogle will make Carbon available starting to a subset of cyber defenders and will gradually phase access to additional trusted partners through its Fairwind Program, collecting feedback and iterating on guardrails before broader developer/business/consumer availability.
VERIFIABLE FACTS
Every value stays attached to a source and date.
RELEASE · Announcement dateDEVELOPER CLAIM
Announced on 2026-10-28 as Gemini 4 Carbon.
RELEASE · Relation to Gemini 4 ArgonDEVELOPER CLAIM
Built on Gemini 4 Argon and announced two weeks after Argon.
CAPABILITIES · Cybersecurity capabilitiesDEVELOPER CLAIM
Described as having state-of-the-art cybersecurity capabilities with significant gains across tasks important for cyber defense (finding, exploiting, and patching vulnerabilities).
BENCHMARKS · Artificial Analysis Intelligence Index (v5.0) rankingDEVELOPER CLAIM
Ranks #1 on Artificial Analysis Intelligence Index v5.0 with a score of 59.
BENCHMARKS · Specific cybersecurity benchmark scoresDEVELOPER CLAIM
Reported scores include SEC-bench Pro 88.2% (finding vulnerabilities), ExploitGym 54.8% (exploitation), and CWE-bench 71.0% (patching).
CAPABILITIES · Long-horizon / sustained agentic workflowsDEVELOPER CLAIM
Designed for much harder, long‑horizon engineering problems and sustained agentic workflows; internal teams report running the model for days to persist through dead ends and handle complex, multi-step engineering tasks (e.g., root-causing and patching subtle bugs across large codebases and low-level systems).
AVAILABILITY · Initial availability / rollout planDEVELOPER CLAIM
Google will make Carbon available starting to a subset of cyber defenders and will gradually phase access to additional trusted partners through its Fairwind Program, collecting feedback and iterating on guardrails before broader developer/business/consumer availability.
SAFETY · Pre-release government testingDEVELOPER CLAIM
Google is actively engaged in the U.S. government’s voluntary process for pre-release model testing for Carbon.
WHAT CHANGED