Google Vertex AI (gateway) serves both Gemini and Anthropic's Claude models under Google Cloud IAM, and it is the enterprise path to Gemini for Atlas. Context varies by model, with 1M on Gemini 3.1 Pro, and pricing is passthrough at Google Cloud rates, with Gemini 3.1 Pro at $2 / $12 per Mtok. Atlas ships a dedicated google-vertex-anthropic route, so Claude on GCP is a first-class provider rather than a workaround. The cost is real setup: a project, a region, and enabled APIs.
What is Google Vertex AI best at inside Atlas?
Google Vertex AI (gateway) is best inside Atlas at running Claude and Gemini side by side under one GCP project, billed to one account. Vertex serves claude-opus-4-1 and claude-haiku-4-5 alongside Gemini 3.1 Pro at $2 / $12 per Mtok, all under Google Cloud IAM.
The pairing is what makes Google Vertex AI (gateway) useful for an Atlas setup. Point the build agent at Gemini 3.1 Pro with its 1M window when a task needs to hold a whole subsystem in context, and set claude-haiku-4-5 as the cheap auxiliary model for titles and summaries, all on the same GCP project and the same bill. Atlas ships a dedicated google-vertex-anthropic route, so Claude on Vertex is a first-class provider rather than a workaround, and Atlas lets you switch the active model and provider on the fly with favorites and recents. Both models appear in `/models`.
How does Atlas authenticate to Google Vertex AI?
Google Vertex AI (gateway) uses service-account authentication via `GOOGLE_APPLICATION_CREDENTIALS` rather than a long-lived API key, which most security teams require. Set `GOOGLE_VERTEX_PROJECT=my-project` and `GOOGLE_VERTEX_LOCATION=us-central1`, and Atlas also falls back to `GOOGLE_CLOUD_PROJECT` and `GCP_PROJECT`.
The absence of a long-lived API key is the whole point for a regulated team. With Google Vertex AI (gateway), Atlas authenticates by pointing `GOOGLE_APPLICATION_CREDENTIALS` at a service account JSON, or by running `gcloud auth application-default login` on a developer laptop. Rotation, revocation, and audit all run through Google Cloud IAM rather than through a key pasted in a shell profile. Once credentials resolve, confirm what Atlas actually sees with `atlas models google-vertex`. Layer Atlas's own controls on top: every Atlas tool call is permission-gated against allow, ask, and deny rules before it runs.
How much does Gemini 3.1 Pro cost on Vertex AI?
Gemini 3.1 Pro costs $2 / $12 per Mtok on Google Vertex AI (gateway), and Vertex pricing is passthrough at Google Cloud rates. The 1M context on Gemini 3.1 Pro is the largest window Vertex exposes to Atlas, and the input side at $2 is the cheaper half by six times.
The $2 / $12 split on Gemini 3.1 Pro shapes how you should drive Atlas on Google Vertex AI (gateway). Input is cheap, so feed it generously: Atlas searches code with hybrid semantic and keyword retrieval fused by reciprocal rank fusion and indexes code by AST declarations using tree-sitter, and pushing a wide retrieval result into a 1M window costs $2 per million tokens. Output at $12 is six times that, and every unified diff Atlas computes is output. For long refactors, plan first in the read-only plan agent, then let the build agent generate diffs against an already-loaded context rather than regenerating it.
What are the real tradeoffs of running Atlas on Vertex AI?
Google Vertex AI (gateway) costs setup time: a GCP project, a region such as us-central1, and enabled APIs before a single token flows. Region availability for the newest models also lags the direct Gemini API by weeks, so Vertex is not where new Gemini releases land first in 2026.
Both tradeoffs on Google Vertex AI (gateway) are worth naming plainly. The setup cost is real: a project, `GOOGLE_VERTEX_LOCATION=us-central1` or another region, enabled APIs, and a service account, before Atlas can send a single token. An API key provider gets you running in one command. The second cost is timing. Region availability for the newest models lags the direct Gemini API by weeks, so if a new Gemini release matters to you on day one, Vertex is the wrong door. What you buy for those costs is IAM, auditability, and one GCP bill covering both Gemini and Claude.
When should you pick the direct Gemini API instead of Vertex AI?
Pick the direct Gemini API over Google Vertex AI (gateway) when you want the newest model on release day, or when a GCP project is more setup than the task deserves. Vertex region availability for the newest models lags the direct Gemini API by weeks in 2026.
The decision on Google Vertex AI (gateway) is governance versus speed of setup. If your organization already runs on GCP and security requires service-account authentication via `GOOGLE_APPLICATION_CREDENTIALS` rather than a long-lived API key, Vertex is the answer and the setup cost is one you were paying anyway. If you are a solo developer who wants Atlas talking to Gemini this afternoon, the direct API is faster. Because Atlas lets you switch the active model and provider on the fly with favorites and recents, running both is normal: Vertex for work code, direct for experiments, both starred in `/models`.
Setup
- 01Set `GOOGLE_VERTEX_PROJECT=my-project` and `GOOGLE_VERTEX_LOCATION=us-central1` (Atlas also falls back to `GOOGLE_CLOUD_PROJECT` and `GCP_PROJECT`).
- 02Point `GOOGLE_APPLICATION_CREDENTIALS` at your service account JSON, or run `gcloud auth application-default login`.
- 03Confirm what resolved: `atlas models google-vertex`.
- 04Select a Gemini or Claude model from `/models`; for Claude on GCP, Atlas routes through its google-vertex-anthropic provider.
Frequently asked questions
- how to configure atlas for google vertex ai
- Set `GOOGLE_VERTEX_PROJECT=my-project` and `GOOGLE_VERTEX_LOCATION=us-central1`, point `GOOGLE_APPLICATION_CREDENTIALS` at your service account JSON or run `gcloud auth application-default login`, then confirm with `atlas models google-vertex`.
- how much does gemini 3.1 pro cost on vertex ai
- Gemini 3.1 Pro costs $2 / $12 per Mtok on Google Vertex AI, which is passthrough at Google Cloud rates. Its context window is 1M tokens, the largest Vertex exposes to Atlas.
- can i run claude on google vertex ai with atlas
- Yes. Vertex serves claude-opus-4-1 and claude-haiku-4-5 alongside Gemini, and Atlas ships a dedicated google-vertex-anthropic route, so Claude on GCP is a first-class provider rather than a workaround.
- does vertex ai need an api key
- No. Google Vertex AI uses service-account authentication via `GOOGLE_APPLICATION_CREDENTIALS` rather than a long-lived API key, which most security teams require. A developer laptop can use `gcloud auth application-default login` instead.
- vertex ai vs the direct gemini api for a coding agent
- Vertex gives you Google Cloud IAM, service-account auth, and one bill covering Gemini and Claude. The direct Gemini API is faster to set up, and region availability for the newest models on Vertex lags it by weeks.
- what environment variables does atlas use for vertex ai
- Atlas reads `GOOGLE_VERTEX_PROJECT` and `GOOGLE_VERTEX_LOCATION`, falling back to `GOOGLE_CLOUD_PROJECT` and `GCP_PROJECT`. Credentials come from `GOOGLE_APPLICATION_CREDENTIALS` or from `gcloud auth application-default login`.
- why is my vertex ai model missing in atlas
- Vertex requires a GCP project, a region selection, and enabled APIs before a single token flows, and newest-model region availability lags the direct Gemini API. Run `atlas models google-vertex` to see exactly what resolved.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas vs Trae: AI Coding Agents in 2026
Atlas and Trae offer distinct AI coding agent experiences in 2026. Atlas provides terminal-native control and local data privacy, while Trae offers a full IDE with SOLO Builder and cloud tasks.
Atlas for Expo: Terminal-Native AI Coding for expo-router and Config Plugins in 2026
Atlas is a terminal-native AI coding agent for Expo apps in 2026, covering expo-router file routes, config plugins, and EAS build profiles with diff-first review.
Atlas for Go in 2026
Atlas, the terminal-native AI coding agent, empowers Go developers in 2026 with intelligent code understanding, safe refactoring, and robust testing capabilities.
Atlas for dbt: Terminal-Native AI Coding in 2026
Atlas is a terminal-native AI coding agent for dbt. Read the ref() DAG, convert a table model to incremental, run dbt build against dev, and add tests in 2026.
Atlas vs Mistral Vibe for Code: Terminal AI Coding Agents in 2026
Compare Atlas and Mistral Vibe for Code in 2026. Atlas offers terminal-native TUI, explicit diffs, and BYO models. Mistral Vibe for Code provides a four-model stack, multi-platform access, and EU data sovereignty.
Atlas for Polars: Terminal-Native AI Coding in 2026
Atlas is a terminal-native AI coding agent for Polars. Build LazyFrame chains, push scan_csv predicates into the reader, and read explain() plans in 2026.
Atlas vs Kilo Code: Terminal AI Coding Agents in 2026
Atlas and Kilo Code in 2026: Compare terminal-native TUI vs VS Code/JetBrains agents. Evaluate pricing, code safety, deployment, and model routing for AI coding.
Atlas vs Tabby: Terminal AI Coding Agents in 2026
Atlas and Tabby comparison for 2026. Atlas offers terminal-native TUI with permission-gated tools and diff review. Tabby provides self-hosted GPU completion and a cloud agent.