Google launches Gemini 4 Argon with Workspace integrations for multi-step AI workflows

Google ships Gemini 4 Argon

Google has made Gemini 4 Argon available on Google Cloud, positioning the model for lengthy, multi-step work where results must be precise and possible to verify. Alongside the model release, Gemini Apps can connect to a range of Google Workspace services, including Gmail, Docs, Drive, Tasks, Keep and Calendar. The details come from a dev.to post summarizing Google Cloud's announcement.

Two different delivery paths

An important nuance in the announcement is that the model and the Workspace experience arrive through separate channels. Argon itself is accessible via the Gemini API under the identifier gemini-4-argon, as well as in Google AI Studio and Google Antigravity. The Workspace-connected experience, by contrast, is delivered through Gemini Apps and its integrations rather than through the API directly.

Argon also appears in the Gemini Apps model catalog with tiered usage and daily prompt quotas spanning Pro, Expanded and Ultra plans. What any individual user can actually do with the model depends on their account, license and configuration.

Built for verifiable, structured work

According to the dev.to write-up, Google highlights three capabilities that suit multi-step tasks: code execution, file handling and structured outputs. Structured outputs are the standout for business use. Instead of an unstructured summary, a team can request a defined set of fields, such as task, owner, due date and status, then feed that into an existing process after review. Code execution and file handling broaden the kinds of developer workflows that can be assembled on top of the API.

The post sketches a realistic pattern for connected workflows: gather relevant material spread across email, documents, files, notes and calendars; have Gemini synthesize it into drafts, outlines, task lists or slide material; and keep a person verifying the result before anything goes out. It singles out client commitments, quantities, dates and pricing as areas where human review is not optional.

Introductory pricing doubles after November

Google's introductory API pricing is $2 per million input tokens and $10 per million output tokens through November 30, 2026. After that date, the stated standard rates are $4 per million input tokens and $20 per million output tokens. Cached input tokens carry a 95 percent discount during the introductory period.

The dev.to post cautions that these figures apply to the Google Cloud model offering, not necessarily to every Gemini Apps or Workspace plan. It suggests separating two questions when estimating cost: which license provides the intended user experience, and whether a custom workflow built on the Gemini API is needed at all. Everyday drafting and organization may be covered by the connected apps, while structured outputs, file processing or connections beyond the documented Workspace experience point toward a custom build.

Data handling needs attention

Google notes that conversations and content in Gemini Apps may be subject to data-use policies and review for some accounts, depending on license and configuration, and that some Workspace services have live-chat limitations. Before connecting sensitive business material, the post recommends that administrators and users review their settings, the applicable terms and the available controls for their account.

Why it matters

Argon signals Google pushing Gemini beyond single-prompt answers toward orchestrated work that draws on a company's existing information. For organizations already standardized on Workspace, the combination offers a direct route from scattered emails, documents and calendars to drafts, task lists and schedules, without immediately replacing the tools they already use. The doubling of API prices after November 30, 2026 is a planning consideration for anyone building on the model, and the tiered quotas in Gemini Apps mean the experience will vary by plan. The constraint is equally plain: the quality of any connected workflow still depends on the quality and accessibility of the underlying files, and on human review at the points where real commitments are made.