Skip to main content

Kaer NewsSoftware

Google releases Gemini 4 Carbon with swarm agents and lower costs

Google launched its strongest AI model today for developers at two dollars per million input tokens.

4 min read
Google releases Gemini 4 Carbon with swarm agents and lower costsPhoto: Google

Key points

Google released Gemini 4 Carbon to developers and enterprises today with swarm agents and pricing at two dollars per million input tokens.

Google released its newest artificial intelligence model called Gemini 4 Carbon to developers and enterprises on November 10 2026. The company says this is its strongest model yet with frontier performance in cybersecurity and long-horizon capabilities for complex multi-step problems. Engineers can now build with the model using the paid tier of the API in Google AI Studio and Android Studio to access these advanced features.

The new model includes a preview of a feature called /swarm on Google Antigravity which turns a single prompt into a coordinated team of agents. Developers can assign the model to roles demanding deep reasoning like architectural design while pairing it with faster lighter models across parallel isolated workspaces. This allows teams to track agents in real time during complex software engineering tasks.

New swarm feature previewed

Google says the model pushes the frontier even further than Gemini 4 Argon which launched last month and remains the ideal choice for most software engineering tasks. Carbon is built for tasks that test the limits of a models intelligence like complex multi-step coding and open-ended scientific research and discovery using more tokens than Argon to explore thoroughly.

The company states the model ranks first on three independent benchmarks designed to test model performance in real workflows including CursorBench 4.0 and Perplexitys WANDR. It achieves these results at more than two times lower cost per task compared to the next best external model available on the market today for developers and enterprises.

Google announced an introductory price of two dollars per million input tokens and ten dollars per million output tokens for the model via the API. This pricing is around a fifth the cost of OpenAIs GPT-6 Astra and a quarter the cost of Anthropics Fable 5.2 model which launched yesterday according to reporting from The Verge.

Pricing and availability details

The company is giving ten thousand scientists free access to the model for six months to help with research including protein structure prediction necessary for developing medications. Google has already onboarded thousands of researchers to this initiative which would presumably help train its models in turn while advancing scientific discovery in healthcare and other fields.

Benchmarks cited by Google when it first announced Carbon suggest the model is one of the most powerful on the market across most metrics behind Fable 5.2. The company says the model can help its first users uncover zero-day vulnerabilities in a short span of time following initial release to cybersecurity defenders two weeks ago.

This release marks a decisive return for Google at the very forefront of artificial intelligence research after months where frontier development looked like a two-horse race between Anthropic and OpenAI. The model will roll out to all Google AI Ultra subscribers in the Gemini app and AI Mode in Search starting today on November 10 2026.

Competition and market impact

Google says the model follows strict schemas and formatting to generate polished and precise spreadsheets presentations and reports across legal finance and enterprise operations. It combines frontier visual understanding reasoning and exhaustive retrieval to synthesize grounded insights across large sets of dense documents and data for higher-precision knowledge work and artifact generation.

The company says existing benchmarks do not adequately capture how agents reason natively over dense files with charts images blueprints schematics and tables. Google partnered with Artificial Analysis to launch a new Multimodal Knowledge Agent benchmark that tests a models multimodal performance in agentic workflows where Carbon ranks second while being more than six times lower in average cost per task.

Google showed examples including a team of agents turning historical photos and engineering plates into an interactive three-dimensional exhibition of the Eiffel Tower construction from a single prompt. Another example built an interactive reconstruction of Googles Platform office7 office using seven public City Council planning PDFs with 206 extracted engineering drawing views across ten overnight loops.

Google says the model performs exceptionally well on engineering tasks that require multi-agent collaboration like multi-package migrations and cross-subsystem root-cause investigations. It also supports web app development by running build-and-verify loops and recovers from errors to produce complete working applications with stronger visual design and spatial reasoning for creative coding tasks.

Frequently asked questions

When did Google release Gemini 4 Carbon to the public?

Google released Gemini 4 Carbon to developers and enterprises on November 10 2026 after a limited release to cybersecurity defenders two weeks prior.

What is the price of Gemini 4 Carbon per token?

The model is priced at two dollars per million input tokens and ten dollars per million output tokens during an introductory period until the end of November.

What is the new swarm feature in Antigravity?

The /swarm feature turns a single prompt into a coordinated team of agents that developers can assign to different roles and track in real time.

More from this edition