Databricks adds Gemini 4 Carbon to Unity Gateway a day after Google’s launch

Databricks said on November 10 that Google’s Gemini 4 Carbon is now available on Unity Gateway, its governance layer for enterprise AI. Customers reach the model through Foundation Model APIs.

The release note is short. It says the model is available, names the access route and reminds customers that they are responsible for complying with the applicable terms. It lists no pricing, regions, context window or rate limits for Carbon on Databricks.

What the release note does and does not say

Databricks describes Unity Gateway as the enterprise control plane for governing AI cost, security and access. According to the company’s August 4, 2026 release note, it lets teams control which AI services they can use, route traffic across providers, govern MCP servers and track usage, cost, access and lineage in one place. Carbon is added to that catalog of models alongside others the company has listed this month.

The same release notes show the pace of additions. Anthropic’s Claude Fable 5.2 arrived on Unity Gateway on November 9, and Moonshot AI’s Kimi K3.1 Pro on November 4. Databricks also said that releases are staged, so an account may not see a new feature until a week or more after the initial release date.

What the page leaves out matters for buyers. It does not say which cloud regions serve Carbon, whether it runs on pay-per-token endpoints, or what Databricks charges. The Kimi note, by contrast, named pay-per-token endpoints in us-east-1.

One day after Google opened the model to developers

Google’s own timeline shows how fast this moved. Carbon was unveiled on October 28, 2026, when only a subset of cybersecurity defenders could use it through the Fairwind Program. According to a Gemini API changelog entry we covered, it became generally available to developers on November 10, the same day as the Databricks note. Google’s introductory list price there is $2.00 per million input tokens and $10.00 per million output tokens, running through November 30, 2026. The post-promotion price was not stated.

Those are Google’s direct API rates. The Databricks note does not say whether its customers pay the same, so the figures should not be assumed to carry over.

Who benefits in practice

For companies already routing models through Unity Gateway, the practical change is one more option behind the same permissions, budgets and monitoring. They do not have to open a separate Google contract or key path to test it, assuming their account has the staged release. Databricks also noted this month that AI Functions spending counts toward Unity Gateway budgets, which suggests cost tracking is a central selling point of the product.

On capability, the independent evidence so far is thin. Cognition, the maker of Devin, reported that Carbon scored 40.0% on its FrontierCode 2.0 benchmark, up from 26.3% for Gemini 4 Argon. That is Cognition’s own figure and has not been independently reproduced. On the same board, Cognition listed Claude Fable 5.2 at 49.1%, and Fable is also on Unity Gateway. Teams choosing between the two on Databricks will have to weigh that gap against price: an earlier report of ours put Fable 5.2 at $8 per million input tokens and $40 for output, against Carbon’s introductory $2 and $10 at Google.

Google has also claimed Carbon writes secure code at more than twice Argon’s success rate, based on internal benchmarks. That claim is unverified, and it is not known whether the Fairwind cybersecurity safeguards apply to the version Databricks serves.

The release note does not say what Databricks adds on top of the model, such as guardrails, logging details or data-handling terms. Customers will need the Foundation Model APIs and supported-models pages to confirm limits for their region before planning production use.