What it is
Gemini Enterprise Agent Platform is what Google Cloud has called Vertex AI since 22 April 2026. Google's announcement describes it as the evolution of Vertex AI and says all Vertex AI services will be delivered through it from now on. It bundles Model Garden (more than 200 models, including Gemini, Gemma, Anthropic's Claude family, Grok and DeepSeek), model tuning and serving, and an agent stack: the Agent Development Kit, Agent Runtime, Memory Bank, and identity and gateway controls for agents.
Models are billed per token. According to the pricing page, Gemini 3.8 Flash costs $0.75 input and $3.75 output per million tokens until 31 December 2026, rising to $1.50 and $7.50 on 1 January 2027. Gemini 4 Argon, generally available since 13 October 2026, is $2 and $10 until 30 November, then $4 and $20. Reserved capacity is sold as provisioned throughput in GSUs (generative AI scale units).
Our review
For most buyers this platform is the enterprise way to use Gemini, and a credible way to run Claude without leaving a Google Cloud perimeter. Google's newest models land here first, and Anthropic's follow within days: according to the release notes, Claude Opus 5.5 arrived in Model Garden on 22 September, Sonnet 5.5 on 28 September and Haiku 5.5 on 7 October. One project, one IAM policy and one invoice cover both families, which is the main reason to choose it over calling each provider directly.
Budgets need to account for two cliffs. Gemini 4 Argon reached general availability on 13 October at $2 input and $10 output per million tokens, but that is promotional pricing delivered as 50% credits back, and it doubles to $4 and $20 on 1 December. Gemini 3.8 Flash doubles from $0.75 and $3.75 to $1.50 and $7.50 on 1 January 2027. Regional endpoints, which you need when data must stay in a region, add 10% to every token, and the pricing page says a tuned model's endpoint costs 1.5 times its base model, not the same price as earlier versions of this review stated. Batch and flex runs are half price.
The data terms are sound but have exceptions worth reading. Google says it will not train on customer data without permission, and request logging is off by default. Gemini still caches prompts in memory for up to 24 hours unless an admin disables it, Grounding with Google Search keeps query logs for three days with no opt-out, and the Interactions API stores conversations unless you set store to false. Teams that have promised zero retention to their own customers need to switch those off deliberately.
The rename is a real cost for engineers. The console no longer says Vertex AI, features have been renamed (Agent Engine is now Agent Runtime), and on 18 September the Python SDK moved agent features out of google-cloud-aiplatform into a separate google-cloud-agentplatform package with new namespaces, a breaking change for existing agent code. Bedrock offers a similar model catalogue on AWS with fewer of these moving parts; what it lacks is first access to Gemini.
Where it earns its place
A platform team at a European insurer already on Google Cloud
Serve Gemini and Claude to internal apps with data kept in the EU
Use the EU multi-region endpoint under existing IAM and VPC Service Controls; every token costs 10% more than global, so Claude Sonnet 5.5 is $2.20 in and $11 out per million. Disable the implicit prompt cache to reach zero retention.
A backend engineer running nightly document classification over 50 million pages
Process a large backlog cheaply without real-time latency
Send the jobs through batch, which halves token prices, on Gemini 3.8 Flash. Budget for the January 2027 doubling of the Flash list price before signing off a yearly cost.
A head of AI at a retailer choosing an agent runtime
Deploy customer-facing agents with memory, tool approvals and live evaluation
ADK agents run on Agent Runtime with Memory Bank and Agent Gateway for approved tools. It saves building the plumbing, but the interfaces are Google's own, so moving to AWS later means a rewrite.
A finance lead at a SaaS company spending about $60,000 a month on model APIs
Get predictable unit costs on Gemini
Spend above $50,000 per 30 days now falls into Standard PayGo Tier 4, and provisioned throughput costs $2.74 per GSU-hour on a one-year commitment against $7.14 for one week. A Flexible Savings Plan can discount the rest.
Before you commit
- Setup
- Enable the API in a Google Cloud project, grant IAM roles and call models through the Gen AI SDK or REST. Expect older guides and Stack Overflow answers to use Vertex AI names.
- Integrations
- Gen AI SDK, REST, the google-cloud-agentplatform 2.x Python SDK for agents, ADK, Agent Gateway and Agent Registry, VPC Service Controls and CMEK on sandboxes.
- Data and privacy
- Google says it does not train on customer data without permission. Prompts are cached in memory for up to 24 hours by default, Search grounding logs are kept three days, and Deep Research keeps sessions for up to seven days; each must be handled separately for zero retention.
- Cost at scale
- One billion input and 200 million output tokens a month on Gemini 3.8 Flash cost about $1,500 on the global endpoint today and about $3,000 from January 2027, plus 10% for a regional endpoint.
- Limits
- Throughput depends on Standard PayGo spend tier, with Tier 4 above $50,000 per 30 days; Priority PayGo and provisioned throughput buy guaranteed capacity. Long-context prompts above 200,000 tokens are billed at higher rates on all tokens.
What's good
- First access to Gemini, plus Claude Opus, Sonnet and Haiku 5.5 on the same bill
- More than 200 models in Model Garden
- Batch and flex inference at half price
- No training on customer data without permission; request logging off by default
- Provisioned throughput with terms from one week to one year
What's not
- Gemini 4 Argon doubles on 1 December 2026 and 3.8 Flash on 1 January 2027
- Regional endpoints cost 10% more and tuned endpoints 1.5 times base
- 24-hour prompt cache and Search grounding logs are on unless you change settings
- Python SDK split in September 2026 broke existing agent code
- Agent features tie you to Google-specific APIs
Best for
Teams on Google Cloud, or anyone standardising on Gemini, who want enterprise controls and Claude under one account.
Alternatives
Google Agent Platform (Vertex AI): common questions
- Is Google Agent Platform (Vertex AI) worth it?
- Gemini lands here first and Claude shares the invoice, but two promotional prices end within ten weeks and the rename has split the documentation in two. We rate it 4.2 out of 5.
- How much does Google Agent Platform (Vertex AI) cost?
- Per 1M tokens, e.g. Gemini 3.8 Flash $0.75 in / $3.75 out (to 31 Dec 2026, then $1.50 / $7.50), Gemini 4 Argon $2 / $10 (to 30 Nov 2026, then $4 / $20), Gemini 3.1 Pro Preview $2 / $12, Claude Sonnet 5.5 $2 / $10. Non-global endpoints cost 10% more; batch is half price. Provisioned throughput $2.74 to $7.14 per GSU-hour by term.
- What are the downsides of Google Agent Platform (Vertex AI)?
- Gemini 4 Argon doubles on 1 December 2026 and 3.8 Flash on 1 January 2027; regional endpoints cost 10% more and tuned endpoints 1.5 times base; 24-hour prompt cache and Search grounding logs are on unless you change settings; python SDK split in September 2026 broke existing agent code; agent features tie you to Google-specific APIs.
- Who is Google Agent Platform (Vertex AI) best for?
- Teams on Google Cloud, or anyone standardising on Gemini, who want enterprise controls and Claude under one account.
- How much does Google Agent Platform (Vertex AI) cost for a team or in production?
- One billion input and 200 million output tokens a month on Gemini 3.8 Flash cost about $1,500 on the global endpoint today and about $3,000 from January 2027, plus 10% for a regional endpoint.
- Is Google Agent Platform (Vertex AI) safe for company data?
- Google says it does not train on customer data without permission. Prompts are cached in memory for up to 24 hours by default, Search grounding logs are kept three days, and Deep Research keeps sessions for up to seven days; each must be handled separately for zero retention.
- What are the best Google Agent Platform (Vertex AI) alternatives?
- The closest alternatives we have reviewed are Amazon Bedrock, Together AI, Baseten and OpenRouter.
Answers checked Oct 23, 2026.
Make Google Agent Platform (Vertex AI)? Show this rating on your site
