Cloud, Data and AI: Gemini Code Assist in 2026: Pricing, Cloud Integration & How It Compares

Introduction

Search for Gemini Code Assist pricing today and most of the results are wrong. They describe a generous free tier, 6,000 daily completions, and a Gemini CLI you can sign into with a personal Google account. All of that ended on 18 June 2026.

What is left is a genuinely strong enterprise product with a narrower shape than it had a year ago — and a cloud story that runs opposite to its competitors. Where OpenAI's Codex can now serve inference from Amazon Bedrock, Azure OpenAI or an Oracle commitment, and Claude Code runs across Bedrock, Vertex AI and Microsoft Foundry, Gemini remains a Google Cloud product. That is not a flaw. It is a trade: less portability, more depth inside one estate.

This guide covers what Code Assist actually is after the split, what Standard and Enterprise really cost, what integration means from each of the four major clouds, and how it compares to the rest of the field.

In this article

The June 2026 split: Code Assist vs Antigravity

This is the single most important thing to understand before you read another pricing table, so it goes first.

At Google I/O in May 2026, Google announced that its developer tools were being unified into a multi-agent platform called Antigravity, with Antigravity CLI replacing the Gemini CLI. On 18 June 2026, the Gemini Code Assist IDE extensions and Gemini CLI stopped serving requests for three tiers: Gemini Code Assist for individuals, Google AI Pro, and Google AI Ultra. The "Login with Google" option for those surfaces was removed.

What survived, unchanged, is the paid enterprise line. Google's deprecation page is explicit: organisations on a Gemini Code Assist Standard or Enterprise subscription are not affected, and keep access to both the IDE extensions and Gemini CLI with continued model updates.

So in 2026 there are two products, and conflating them produces bad decisions:

	Gemini Code Assist (Standard / Enterprise)	Antigravity
Who it is for	Teams and organisations with a Google Cloud project	Individual developers, including former free-tier users
Surfaces	IDE extensions (VS Code, JetBrains, Android Studio), Gemini CLI, Cloud Workstations, Cloud Shell Editor	Antigravity editor and Antigravity CLI (`agy`)
Billing	Per-user licence, monthly or annual commitment, bought through Google Cloud	Consumer Google AI subscription tiers plus AI credits
Enterprise controls	Data governance, indemnification, private code indexing, usage metrics	Consumer-grade
Source availability	Gemini CLI is open source	Antigravity CLI is not open source

Practical consequence. If you are a solo developer or a two-person startup, Gemini Code Assist is no longer aimed at you — Antigravity is, and its quota structure has been revised several times through 2026. If you are a team of ten or more buying seats, Code Assist Standard or Enterprise is the product, and none of the free-tier headlines apply.

What Gemini Code Assist does

In the IDE

The extension runs in VS Code, JetBrains IDEs, Android Studio, Cloud Workstations and the Cloud Shell Editor. You get code completion as you type, generation from natural-language comments, unit-test generation, debugging help, and chat grounded in the open file and project. Standard subscribers on the Preview release channel get access to Gemini 3 with a one-million-token context window — the largest in the category, and the reason Code Assist handles large single-file and cross-file reasoning well.

Agent mode and Gemini CLI

Agent mode is a multi-step reasoning agent inside Code Assist chat that plans and executes multi-file tasks rather than answering one prompt at a time. Alongside it, the licence includes Gemini CLI, the open-source terminal agent — code understanding, file manipulation, command execution and troubleshooting from the shell. For enterprise licence holders, this is the surface that matters most for automation.

Code customization: the Enterprise differentiator

This is what you are actually buying when you move from Standard to Enterprise. Code customization connects Code Assist to your private repositories and builds a private index. As you type, it searches that index for code similar to what you are writing and folds the matches into the prompt, so suggestions reflect your internal libraries, private APIs and house style rather than generic public-code patterns. The index is refreshed every 24 hours.

The constraints are specific and worth knowing before you plan a rollout:

Repositories: github.com, gitlab.com, bitbucket.org, plus self-hosted GitHub Enterprise, GitLab Enterprise and Bitbucket Data Center — including instances on private networks. GitHub Enterprise Cloud IP restrictions are not supported.
Regions: the underlying Developer Connect connection is supported only in us-central1, europe-west1 and asia-southeast1. This is a hard constraint on data-residency planning.
Languages indexed: C, C++, C#, Go, Java, JavaScript, Kotlin, PHP, Python, Rust, TypeScript, plus Markdown documentation. Everything else is ignored by the index.
IDE versions: VS Code extension 2.18.0+ or the IntelliJ plugin, plus Cloud Workstations and Cloud Shell Editor.

Code review agent

Google also ships a Gemini code review agent that comments on pull and merge requests. Like code customization, it connects through Developer Connect and supports GitHub, GitHub Enterprise, GitLab, GitLab Enterprise, Bitbucket Cloud and Bitbucket Data Center, including private-network instances. Note what is absent from that list — we come back to it in the Azure section.

Gemini Cloud Assist

Distinct from Code Assist, and bundled in part with the Enterprise tier, Cloud Assist is the assistant embedded across the Google Cloud console — troubleshooting, cost and architecture guidance, BigQuery data insights, code completion inside Cloud Run. Enterprise also pulls in Gemini in Apigee for API design and Gemini in Application Integration. This bundle is the real argument for Enterprise in a GCP-heavy organisation, and it is worth nothing at all if you run on AWS.

Pricing: Standard, Enterprise, and the real bill

Google prices Code Assist as a per-user licence with a discount for annual commitment. Published rates:

Tier	Monthly commitment	Annual commitment	What it adds
Standard	$22.80 / user / mo	$19.00 / user / mo	IDE completion and generation, chat, agent mode, Gemini CLI, Gemini 3 on the Preview channel, usage metrics dashboard, enterprise data governance and indemnification.
Enterprise	$54.00 / user / mo	$45.00 / user / mo	Everything in Standard plus code customization on private repos, full codebase awareness, Gemini in Apigee and Application Integration, additional Cloud Assist features.

The annual commitment saves roughly 17%. Google's own pricing page quotes these as hourly rates — Standard at about $0.0312/hour and Enterprise at about $0.0740/hour — which multiply out to the monthly figures above across a 730-hour month. There is a 30-day free trial covering up to 50 users, which is unusually generous next to the single-seat trials most competitors offer.

What a team actually pays

Team size	Standard (annual)	Standard (monthly)	Enterprise (annual)	Enterprise (monthly)
5 developers	$95 / mo	$114 / mo	$225 / mo	$270 / mo
25 developers	$475 / mo	$570 / mo	$1,125 / mo	$1,350 / mo
100 developers	$1,900 / mo	$2,280 / mo	$4,500 / mo	$5,400 / mo

The budgeting advantage nobody mentions

Here is where Code Assist quietly wins on cost predictability. It is a flat per-seat licence with published quotas. It does not meter your agent's token consumption and bill you for overage.

Compare that to the rest of the category. OpenAI's own documentation puts average Codex spend at roughly $100–$200 per developer per month. Anthropic's puts Claude Code on API billing at $150–$250. GitHub moved Copilot to usage-based AI Credits in June 2026. A hundred developers on Code Assist Enterprise costs $4,500 a month and you know that number in advance; a hundred developers running token-metered agents all day is somewhere between $10,000 and $20,000 and you will not know which until the second invoice.

The honest counterweight: you are buying a different intensity of tool. Code Assist is completion-plus-agent-mode; Codex and Claude Code are long-horizon autonomous agents. Predictable cost partly reflects a bounded workload.

What is not included in the seat price

Gemini API or Vertex AI usage for anything you build yourself — that is separate, per-token billing.
Cloud Workstations compute, if you run developer environments there.
Apigee and Application Integration platform costs; the Enterprise tier includes the Gemini assistance in those products, not the products themselves.
Egress and networking for private-repo indexing through Developer Connect.

Pros and cons

Strengths	Trade-offs
Flat per-seat pricing with no token metering — by far the easiest AI coding tool in this category to budget for.	No individual or free tier since June 2026. Solo developers and tiny teams are pushed to Antigravity, a different product with a volatile quota history.
1M-token context window with Gemini 3, the largest in the category, so large-codebase reasoning holds up.	Gemini runs only on Google Cloud. No Bedrock route, no Azure route, no OCI route — unlike Codex and Claude Code.
Code customization grounds suggestions in your private repositories across GitHub, GitLab and Bitbucket, including self-hosted and private-network instances.	Code customization is region-limited to three Developer Connect regions and indexes only eleven languages plus Markdown.
Enterprise data governance, no training on your prompts and responses, and indemnification for code suggestions.	Enterprise is roughly 2.4× the price of Standard, and much of what it adds (Apigee, Cloud Assist, Application Integration) is worthless outside Google Cloud.
Deep integration across the Google Cloud console, BigQuery, Cloud Run and Apigee — no other assistant matches this inside GCP.	Azure DevOps Repos is not a supported Developer Connect source, which is a real gap for Microsoft-stack teams.
Gemini CLI is open source, and a 30-day, 50-user trial makes proper evaluation cheap.	The Antigravity transition has left enterprise users asking publicly how quickly new Gemini models reach the Code Assist CLI. Check model currency during your trial.

Cloud integration: Google Cloud, AWS, Azure, OCI

Start with the fact that governs everything else in this section.

Gemini frontier models are exclusive to Google Cloud. When AWS added Google DeepMind models to Amazon Bedrock on 10 June 2026, what arrived was the Gemma 4 open-weight family — Gemma 4 31B, 26B-A4B and E2B, under Apache 2.0, with a 256K context window. Useful models, and genuinely helpful for AWS-native inference workloads, but they are not Gemini and they do not give you Gemini Code Assist. There is no equivalent of Codex-on-Bedrock or Claude-Code-on-Bedrock for Gemini.

So "integrating Gemini Code Assist with AWS" cannot mean moving inference into AWS. It means something narrower and still useful: the licence and the model stay on Google Cloud, while the code, the CI and the deployment targets live wherever you like.

Google Cloud: the native case

On GCP, Code Assist is not a coding plugin that happens to be from Google — it is woven through the platform:

Console-wide assistance via Gemini Cloud Assist: troubleshooting, architecture and cost guidance against your actual project.
BigQuery data insights — generated query libraries derived from your table metadata.
Cloud Run code completion inside the source editor, context-aware to the service you are editing.
Cloud Workstations and Cloud Shell Editor — Code Assist, including code customization, works in browser-based environments, so you can give contractors a fully governed workstation with no code on local disk.
Apigee and Application Integration (Enterprise) — API design grounded in your API hub as "enterprise truth".
IAM and quota handled through the same Google Cloud project that enables the Gemini for Google Cloud API, so licences, access and billing sit in one place.

If your data gravity is BigQuery and your services run on Cloud Run or GKE, nothing else in this article competes.

Amazon Web Services: what works and what does not

What works. The IDE extension runs on any laptop regardless of what you deploy to. Your repository can live on GitHub or Bitbucket and be indexed by code customization. The code review agent comments on your pull requests. Gemini CLI and MCP let the agent reach AWS context — point it at AWS's published MCP servers so it can read CloudFormation state, CloudWatch logs or your own platform APIs rather than guessing. For CI, you can run Gemini CLI inside CodeBuild and authenticate with workload identity federation instead of a service-account key.

What does not. No Bedrock provider route, so inference leaves your AWS boundary and does not draw down an AWS commit. No IAM-native access control over model calls, no CloudTrail record of them, no PrivateLink path. Gemini Cloud Assist has nothing to say about your AWS estate. If your requirement is "model calls must stay inside our AWS account", Code Assist cannot meet it and Amazon Q Developer or a Bedrock-routed agent can.

Microsoft Azure: the Developer Connect gap

The AWS picture applies to Azure with one extra problem worth flagging, because it catches teams mid-rollout.

Developer Connect — the service both code customization and the code review agent depend on — supports GitHub, GitHub Enterprise, GitHub Enterprise Cloud, GitLab, GitLab Enterprise and Bitbucket in both Cloud and Data Center flavours. Azure DevOps Repos is not on that list. If your source of truth is Azure Repos rather than GitHub, the two headline Enterprise features do not reach your code, and you are paying $45–$54 per seat for the Standard feature set.

Workarounds exist and each has a cost: migrate repositories to GitHub Enterprise (which many Microsoft-stack organisations are doing anyway), mirror Azure Repos to a GitHub organisation purely for indexing, or accept Standard-tier value. None of these is free, and all of them should be priced into the business case rather than discovered afterwards.

Otherwise the pattern holds: extensions run in VS Code natively, Gemini CLI runs in Azure Pipelines, MCP connects the agent to Azure context, and workload identity federation from Entra ID to Google Cloud removes static keys from the pipeline.

Oracle Cloud Infrastructure

Of the four, OCI has the thinnest story, and it is worth being direct about that. There is no Gemini route through OCI Generative AI, no Universal Credits path of the kind Oracle opened for OpenAI models and Codex in June 2026, and no Oracle-specific grounding in Code Assist.

What you can do is the same generic pattern — IDE extension on the developer's machine, repositories on GitHub or Bitbucket for indexing, Gemini CLI in OCI DevOps pipelines with federated credentials, MCP servers exposing OCI APIs. For teams running Oracle Database, Java and Fusion workloads, Oracle Code Assist is the natively integrated option and is priced comparably. Realistically, Gemini Code Assist belongs in an OCI shop only when the development teams have already standardised on Google Cloud for something else.

Side-by-side

Capability	Google Cloud	AWS	Azure	OCI
Gemini inference inside your cloud	Yes (Vertex / Gemini Enterprise)	No — Gemma 4 only	No	No
Model calls under your cloud IAM and audit log	Yes	No	No	No
Pay from existing cloud commit	Yes	No	No	No
IDE extension and Gemini CLI	Yes	Yes	Yes	Yes
Private-repo code customization	Yes	Yes (GitHub / GitLab / Bitbucket)	Only if repos are not in Azure Repos	Yes (GitHub / GitLab / Bitbucket)
Console and platform grounding	Deep (Cloud Assist, BigQuery, Apigee)	Via MCP only	Via MCP only	Via MCP only
Keyless CI authentication	Native	Yes (WIF)	Yes (WIF)	Yes (WIF)

Running it in non-Google CI with workload identity federation

If you take one operational idea from this article, take this one. Do not put a Google service-account JSON key into AWS Secrets Manager, Azure Key Vault or an OCI vault. Use workload identity federation, which lets a workload running in another cloud exchange its own identity token for short-lived Google Cloud credentials. No key to rotate, no key to leak, and the trust relationship is auditable on both sides.

The AWS shape, for a CodeBuild job that runs Gemini CLI:

# 1. Create a workload identity pool and an AWS provider
gcloud iam workload-identity-pools create ci-pool \
  --location="global" --display-name="CI pool"

gcloud iam workload-identity-pools providers create-aws aws-provider \
  --location="global" \
  --workload-identity-pool="ci-pool" \
  --account-id="123456789012" \
  --attribute-mapping="google.subject=assertion.arn"

# 2. Let one specific IAM role impersonate a Google service account
gcloud iam service-accounts add-iam-policy-binding \
  gemini-ci@PROJECT_ID.iam.gserviceaccount.com \
  --role="roles/iam.workloadIdentityUser" \
  --member="principalSet://iam.googleapis.com/projects/PROJECT_NUMBER/\
locations/global/workloadIdentityPools/ci-pool/attribute.aws_role/\
arn:aws:sts::123456789012:assumed-role/CodeBuildGeminiRole"

Then generate a credential configuration file, ship it with the build (it contains no secret), and point GOOGLE_APPLICATION_CREDENTIALS at it. The same pattern works with an Entra ID OIDC provider on Azure and with OCI. Restrict the binding to one exact role or principal rather than a whole account — a broad principalSet is the most common misconfiguration here. More on hardening agents in pipelines in our DevOps coverage.

Gemini Code Assist vs. the competition, with prices

Published list prices gathered in September 2026. Every vendor here has changed pricing at least once in the past year.

Tool	Shape	Entry price	Team / enterprise	Billing model
Gemini Code Assist	IDE extension, agent mode, Gemini CLI, 1M context	No individual tier	$19–22.80 Standard; $45–54 Enterprise	Flat per seat, no token metering
GitHub Copilot	IDE completion plus agents, deepest GitHub integration	$0 Free / $10 Pro	$19 Business, $39 Enterprise, $100 Max	Seat plus usage-based AI Credits
ChatGPT Codex	Agentic harness across CLI, IDE, desktop, cloud	$0 Free / $8 Go / $20 Plus	$20–25 per user; Pro $100–$200	Token credits; ~$100–200 per dev per month in practice
Claude Code — our review	Terminal-first agent, IDE and CI surfaces	$20/mo Pro	Max $100 / $200; Team and Enterprise seats	Subscription or per-token; ~$150–250 per dev per month on API
Cursor	AI-native editor with multi-file agent	$0 Hobby / $20 Pro	Pro+ $60, Ultra $200, Teams ~$40/seat	Seat plus usage pool
Amazon Q Developer	IDE plugin with AWS-native tooling (IAM, CloudFormation, Java upgrades)	Free tier, capped agentic requests	$19/user Pro	Flat per seat
Oracle Code Assist	IDE assistant tuned for Oracle DB, Java, Fusion	—	~$19–$39/dev indicative list	Per seat, negotiable against Universal Credits
Tabnine	Enterprise assistant with air-gapped deployment option	—	~$39–$59/user, annual commitment	Per seat; the realistic pick under strict zero-retention rules

Three things the table does not show

Gemini Code Assist is now the expensive entry point. A year ago it was the generous free option. Today, Standard at $19–22.80 is more expensive per seat than Copilot Business at $19 with a far larger install base, and Enterprise at $45–54 sits above Copilot Enterprise at $39. Google's pricing now has to be justified by 1M context, code customization and Google Cloud depth — not by being cheap.

Flat pricing is a genuine feature, not a consolation. Finance teams are, by late 2026, noticeably tired of AI line items that move 5× month to month. If you are the person who has to forecast this spend, a per-seat licence with published quotas is worth real money in avoided variance, and it is the strongest argument for Code Assist outside Google Cloud.

Cloud portability sorts the field cleanly. Codex runs on Bedrock, Azure OpenAI and against an Oracle commitment. Claude Code runs on Bedrock, Vertex AI and Microsoft Foundry. Gemini Code Assist runs on Google Cloud. If multi-cloud model routing is a requirement in your architecture review, that is the paragraph that decides it.

How to choose for your stack

Google Cloud native. Straightforward — Code Assist Enterprise if you want private-repo grounding and Apigee assistance, Standard if you mainly want strong completion and Gemini CLI. Nothing else integrates as deeply with your console.
AWS-heavy. Amazon Q Developer at $19 for AWS-specific work, plus a portable agent (Codex on Bedrock, or Claude Code) for delegated tasks. Code Assist earns a seat here only if you specifically want the 1M context window or already hold Google Cloud commitments.
Azure or Microsoft-stack. Check where your repositories live first. Azure Repos means the Enterprise features do not reach your code — decide on a migration before you sign, not after.
Oracle estate. Oracle Code Assist for Oracle-specific work; Codex through Universal Credits if you want a frontier agent inside existing procurement. Gemini is the outside option.
Cost predictability is the top constraint. Code Assist wins outright. Flat per-seat, published quotas, no overage surprises.
Solo developer or very small team. Code Assist is not for you any more. Copilot Pro at $10 or Antigravity are the realistic entry points.
Mixed reality, which is most teams. Run a completion tool and an agentic tool. Gemini Code Assist is an excellent completion-and-context tool with real enterprise governance; it is not the strongest long-horizon autonomous agent in this list, and buying it as though it were will disappoint.

Frequently asked questions

Is Gemini Code Assist still free?

No. Starting 18 June 2026, the Gemini Code Assist IDE extensions and Gemini CLI stopped serving requests for the individuals tier and for Google AI Pro and Ultra subscriptions, and the "Login with Google" option was removed from those surfaces. Only paid Standard and Enterprise subscriptions remain. Any guide describing a free tier with thousands of daily completions predates that date.

What happened to Gemini CLI?

For consumer and free-tier users it was replaced by Antigravity CLI. For organisations with a Code Assist Standard or Enterprise licence, Gemini CLI access is unchanged and Google has said it will continue to support it with access to current Gemini models. Verify model currency during your trial if that matters to you.

How much does Gemini Code Assist cost per user?

Standard is $22.80 per user per month on a monthly commitment or $19.00 on an annual commitment. Enterprise is $54.00 monthly or $45.00 annual. There is a 30-day free trial for up to 50 users.

Can I run Gemini Code Assist on AWS or Azure?

You can use the IDE extension and Gemini CLI from any machine and target any cloud, and you can index repositories hosted on GitHub, GitLab or Bitbucket regardless of where you deploy. What you cannot do is serve Gemini inference from AWS or Azure — there is no Bedrock or Azure OpenAI route for Gemini, so model calls leave your cloud boundary and do not draw down your cloud commitment.

Are Gemini models available on Amazon Bedrock?

Not the Gemini frontier models. On 10 June 2026 AWS made Google DeepMind's Gemma 4 open-weight family available on Bedrock — Gemma 4 31B, 26B-A4B and E2B, with a 256K context window on the 31B variant. Those are useful for AWS-hosted inference but they are not Gemini and do not provide Gemini Code Assist.

Does code customization work with Azure DevOps Repos?

No. Developer Connect, which code customization and the code review agent both depend on, supports GitHub, GitLab and Bitbucket — including enterprise and private-network instances — but not Azure DevOps Repos. Teams on Azure Repos should resolve this before purchasing Enterprise licences.

Does Google train on my code?

Google states that prompts and responses from Gemini Code Assist are not used to train the underlying models, and that Standard and Enterprise are covered by enterprise data governance with encrypted input and output. Code customization stores embeddings of your code for indexing rather than plain code. As always, confirm the specific terms in your contract rather than relying on a marketing page.

Gemini Code Assist or GitHub Copilot?

Copilot has the larger ecosystem, the cheaper entry point, and much deeper GitHub and pull-request integration. Code Assist has the larger context window, private-repo code customization, flat non-metered pricing, and unmatched Google Cloud depth. If GitHub is your hub and you want the cheapest credible option, Copilot. If you run on Google Cloud, or you need predictable spend at scale, Code Assist.

References

Pricing and feature details for every product mentioned here change frequently. Confirm current numbers on each vendor's official pricing page before making a purchasing decision. Antigravity's quota structure in particular has been revised several times during 2026 — check Google's current plans page rather than any third-party summary.