Google unveils Gemini 4 Carbon with advanced cybersecurity features

Investing.com -- Google introduced its latest artificial intelligence model on Wednesday, called Gemini 4 Carbon, which the company describes as its most powerful model with enhanced cybersecurity capabilities and long-horizon task performance.

The new model achieved the top ranking on the updated v5.0 Artificial Analysis Intelligence Index with a score of 59. Carbon demonstrated improvements in coding and agentic tasks, reaching 88.2% on SEC-bench Pro for vulnerability detection, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching.

The model follows the release of Gemini 4 Argon, which launched two weeks ago. Google stated that Carbon showed reduced misalignment and fabrication rates in live coding environments with internal teams.

The company is participating in the U.S. government’s voluntary pre-release model testing process. Starting this week, Google will provide Carbon access to select cyber defenders and will gradually expand availability to additional partners through its Fairwind Program. The model will later become available to developers, businesses, and consumers.

Gemini 4 Carbon will be priced at $2 per million input tokens and $10 per million output tokens, matching the introductory price of Argon. Cached input tokens will be priced at 95% off the input token price.

Google reported that internal teams have used Carbon to run multi-day workflows for complex engineering tasks, including identifying and fixing bugs across large codebases. The model gained 89 elo points to reach the #4 position at 1769 on WebDev Arena.

The Chrome team reported that Carbon performed more than 10 times better than another lab’s large frontier model when generating high quality patches. Cloud security platform Wiz also observed Carbon leading on benchmarks for web penetration testing.

In testing, Carbon reduced successful cheating on Terminal-Bench 4.0 to 0.0%, with cheating attempts dropping to 0.4%. The model achieved a 0.0% attack success rate on Gray Swan’s Indirect Prompt Injection Benchmark, resisting all evaluated attack attempts.

Original Article

You May Also Be Interested In

• Alphabet has over $1 trillion in future commitments, leases, purchases
• Align Technology names former P&G healthcare CEO as next leader
• HomeTrust Bancshares names former Bank of America executive as director