Gemini 4 Carbon
Gemini 4 Carbon by Google is a frontier model for days-long agentic engineering and cyber defense: finding, validating and patching vulnerabilities.
Model Overview
Capabilities, design details, and architectural traits
Gemini 4 Carbon - Google's long-horizon model for cyber defense
Gemini 4 Carbon is Google's next frontier model after Gemini 4 Argon, built for sustained agentic engineering and the full cybersecurity defense lifecycle. It is designed to keep working through dead ends on hard, multi-step problems, and it reaches cyber defenders before anyone else.
| Trait | Detail |
|---|---|
| Full defense lifecycle | Finds real vulnerabilities, checks which ones are exploitable, and patches them at the root cause |
| Days-long agentic runs | Google teams have seen it run for days on tasks like root-causing and patching subtle bugs across large codebases and low-level systems |
| Secure code generation | Shows a strong uplift over Argon in generating secure code in Google's internal testing |
| Honest under pressure | Puts methodological caveats into the deliverable, or withholds blocked actions with a clear list of what is still required, instead of fabricating |
| Checks before acting | Gathers the relevant material and flags discrepancies to the user when a task is ambiguous, rather than guessing |
| Stays in scope | Trained against shortcuts and loophole exploitation, with less overreach and less under-service on multi-step plans |
Built on Argon's safety stack
Carbon keeps Argon's layered safeguards: model training, real-time misuse defenses, prompt-injection resilience and chain-of-thought monitoring, with further work on alignment for complex or high-pressure tasks.
Defenders-first rollout
Google is releasing Carbon deliberately. Access starts with a subset of cyber defenders and expands in phases to trusted partners through the Fairwind Program, before reaching developers, businesses and consumers.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Anthropic | Anthropic | |
| Release Date | October 28, 2026 | November 9, 2026 | September 22, 2026 |
| Knowledge Cutoff | - | - | - |
| Context & Limits | |||
| Context Window | 1.0M Best Context Window | 1M | 1M |
| Pricing (per 1M tokens) | |||
| Input Pricing | $2 Best Input Pricing | $8 | $4 |
| Output Pricing | $10 Best Output Pricing | $40 | $20 |
| Modalities | |||
| Inputs | text | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 60.0 | 64.6 Best Intelligence Index | 59.1 |
| Coding Index | - | - | - |
| Agentic Index | - | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from Google
Other models by Google
Gemini 4 Hydrogen
Gemini 4 Argon
Gemini 3.7 Flash
Gemini 3.6 Flash
Top AI Models
Leading alternatives by intelligence score
Claude Fable 5.2
GPT-6 Astra
Muse Spark 1.3
