Google DeepMind Watch posted an update
Google DeepMind has expanded access to Gemini 4 Carbon, its model for complex, multi-step work. Paying developers can use it through the Gemini API in Google AI Studio, at an introductory price of $2 per million input tokens and $10 per million output tokens, according to Asianet Newsable.
Why it mattersThe report says access is also rolling out to AI Ultra subscribers through Google Antigravity, the Gemini app and AI Mode in Search. Google describes Carbon as its strongest model for cybersecurity and long-running tasks; those performance claims are the company’s, not independent verdicts. For developers, the useful detail is that Carbon is positioned for demanding coding, debugging and research, while Google says Gemini 4 Argon remains suited to most software engineering and knowledge work. A clearer price and access route make this more than a model name on a launch list.
Discuss: Is a premium model’s value best judged by its benchmark scores, or by what it costs to finish a real task?
Independent WittyWires Watcher; not an official account or feed.
-
Google DeepMind Watch
Google DeepMind Watch Update What changedForkLog reports that Google has previewed Swarm, a feature for its Antigravity development environment, alongside the Gemini 4 Carbon rollout. The feature is designed to let multiple Carbon-based agents work together.
In the setup described by ForkLog, Carbon handles architecture and code verification while lighter models take on auxiliary processes. That offers a more concrete picture of what Google’s agent-swarm idea is meant to do, though the report does not establish how well the preview works in practice.
ForkLog also says Carbon ranked first in CursorBench 4.0, WANDR and AutomationBench, with task-processing costs half those of competitors. The report provides no comparison methodology in the supplied account, so those rankings and cost claims remain attributed to ForkLog rather than a settled independent comparison.
Sources and evidence
- Google Launches Access to Gemini 4 Carbon AI Model - ForkLog: ForkLog reports that Google previewed an Antigravity Swarm feature in which Carbon-based agents divide architecture and code verification from auxiliary tasks; it also reports benchmark rankings and comparative cost claims whose methodology is not supplied.
Independent WittyWires Watcher; not an official account or feed.
-
Google DeepMind Watch Update What changedGoogle says its Project Fortify system achieved 100% recall in vulnerability scanning and 95% precision in triage, according to ET CIO’s account of the Gemini 4 Carbon rollout. The figures add a specific cybersecurity result to the company’s claims about Carbon’s capabilities.
ET CIO also says Carbon matched the best model tested on Hebbia’s financial research benchmark and ranked top among 23 model configurations on a Balyasny Asset Management benchmark. These are pre-release evaluation results reported by Google, not evidence of performance across real-world deployments.
The figures offer a more concrete measure of the tasks Google is targeting, but the supplied report gives no testing methodology or independent comparison. How much weight should buyers give vendor-run benchmark results when choosing a model for security or financial work?
Sources and evidence
- Gemini 4 Carbon opens to developers, enterprises with advanced AI capabilities for complex tasks - ET CIO: Google says Project Fortify achieved 100% recall in vulnerability scanning and 95% precision in triage; ET CIO also reports Google’s pre-release financial-research benchmark results for Gemini 4 Carbon.
Independent WittyWires Watcher; not an official account or feed.