[DigitalToday reporter Yoonseo Lee (이윤서)] Google has released its AI model Gemini4 Carbon, which focuses on long-running, multi-step tasks, to developers, businesses and general users.

Cryptopolitan, a blockchain media outlet, reported on Monday that Google said the model cut per-task costs to less than half compared with external rival models while recording the best performance across three major agent benchmarks.

The model targets multi-stage tasks such as complex coding and debugging, open scientific research and cybersecurity. In a launch post, Google DeepMind senior vice president Koray Kavukcuoglu said Carbon ranked No. 1 on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench.

Carbon is designed to perform more verification steps before answering. It therefore uses more tokens than Gemini4 Argon, a model in the same family. Google said it plans to keep Argon after releasing it last month, and it still recommends Argon for most software engineering and knowledge work. It said Carbon is used for tasks that test the limits of a model's intelligence.

Google also broadened the benchmark scope. It built a multimodal knowledge agent benchmark with Artificial Analysis to handle dense documents such as charts, blueprints and tables. Carbon ranked No. 2 in the evaluation. Google stressed, however, that Carbon's average per-task cost was more than six times lower than the No. 1 model.

Carbon also performed well in finance-related evaluations. Google said Carbon matched the top model in Hevia's end-to-end financial agent evaluation, and ranked No. 1 among 23 frontier model configurations in Valyasnyi's general financial intelligence benchmark.

Initial API pricing for Carbon and Argon is the same: $2 per 1 million input tokens and $10 per 1 million output tokens. After the introductory period ends on Nov. 30, Carbon's input fee doubles to $4 and its output fee doubles to $20.

Google also set up access for general users. Subscribers to Google AI Ultra, priced at $100 a month, can use Carbon in the Gemini app, Search's AI Mode and Antigravity. Antigravity will also offer a preview of a '/swarm' feature that forms multiple agent teams with a single prompt. Developers can assign Carbon to tasks that require advanced reasoning, such as architecture design or auditing, and delegate parallel tasks to lighter models.

Google also disclosed early use cases in security and life sciences. It said the Wiz team used Carbon to find zero-day vulnerabilities at two major SaaS providers within 2 hours. Google researchers found, on average, 40 times more remote homologs than before across seven protein families with low homology. The data is used for protein structure prediction models such as AlphaFold 3.

Google also expanded support for scientific research. It has been accepting applications for the past four weeks from 10,000 scientists for a program that offers six months of free Argon access on Antigravity. Thousands have already joined, and Google will also provide Carbon access to both current and future participants. Google is providing Argon for general tasks and Carbon for high-difficulty reasoning to broaden its developer and researcher base.