Cloud model platform · Verified August 16, 2026

TrustedRouter vs Google Vertex AI

Vertex AI — renamed the Gemini Enterprise Agent Platform in April 2026 — is Google Cloud's managed platform for Gemini, partner, and open models, with deep compliance and a documented but opt-out-based path to zero data retention. TrustedRouter is a cross-provider gateway whose default realtime path keeps no durable prompt logs and runs in a hardware-attested build you can verify.

Compare details

8decision dimensions

8official sources

1base URL to try TR

Choose Google Vertex AI when

Its operating model is the feature.

Choose Vertex AI (Agent Platform) when committed Google Cloud spend, FedRAMP High or HIPAA scope, Provisioned Throughput on Gemini, grounding in Google Search or Maps, or on-prem deployment via Google Distributed Cloud decides the architecture — and a contractual zero-data-retention checklist meets your privacy bar.

Choose TrustedRouter when

Privacy needs evidence.

Choose TrustedRouter when one API must fail over across Google, Anthropic, OpenAI, and roughly 46 other providers, when realtime prompts must never touch durable storage by default rather than after an opt-out checklist, and when you want to verify the gateway build by remote attestation instead of accepting policy.

DimensionGoogle Vertex AITrustedRouter

DeploymentManaged Google Cloud service; Global + regional endpoints; on-prem via Distributed CloudHosted control plane, source-available (BUSL-1.1), with an attested API path

API surfaceGen AI SDKs; OpenAI-compatible Chat Completions for Gemini + select self-deploys onlyOpenAI Chat Completions and Responses plus Anthropic Messages

Model access200+ curated models: Gemini 3.x, Imagen 4, Veo 3.1, Claude Opus 5, DeepSeek, open models500+ models across ~50 providers: prepaid, BYOK, and direct

RoutingGlobal endpoint, service tiers, Provisioned Throughput; no cross-vendor failoverProvider fallback plus auto, cheap, fast, free, ZDR, E2E, and EU routes with composable privacy and jurisdiction preferences

ObservabilityCloud Logging/Monitoring, audit logs; optional BigQuery request logging (off by default)Metadata analytics and opt-in external broadcast

Prompt contentNo-training terms; 24h cache default; abuse logs up to 90d; ZDR via opt-out checklistNo durable prompt or output logs on realtime inference; batch is opt-in encrypted retention

VerificationIAM, FedRAMP High, HIPAA/BAA, CMEK; no attestation on managed Gemini inferenceLive gateway attestation on three clouds, bound to published source and release evidence

BillingSource token prices; three service tiers, promo expiry dates, GSU commits, spend CUDsPrepaid at provider price + 5.5% ($0.01/M floor), or BYOK

What Vertex AI is today

Since April 22, 2026, Vertex AI has been the Gemini Enterprise Agent Platform: roughly 50 renames at Cloud Next 2026, docs re-homed to docs.cloud.google.com, APIs unchanged, no forced migration. We use the old name because most readers still do.

It is Google Cloud's proprietary managed platform for first-party models (Gemini 3.x, Imagen 4, Veo 3.1), partner models (Anthropic Claude, Mistral, DeepSeek, Qwen, and others), and open models, plus MLOps and agent tooling. Model Garden lists a curated set of 200+ models as of August 2026, and the cadence is fast: Gemini 3.7 Flash went GA August 13, 2026; Claude Opus 5 arrived July 24, 2026. Deployment spans the managed API (Global plus regional endpoints), self-deployed GPU endpoints, and on-prem installs via Google Distributed Cloud.

Where it is genuinely strong

Source pricing and capacity engineering on Google models. Gemini 2.5 Pro is $1.25/M input and $10.00/M output at up to 200K-token context ($2.50/$15.00 above), batch runs about 50% off, and cached input drops to $0.125/M. Provisioned Throughput sells committed capacity from $2.7397 per GSU-hour on a one-year Global commit. No gateway replicates those economics on Google's own models, including us.
Compliance depth. FedRAMP High authorization for Generative AI on Vertex, HIPAA coverage with a BAA, VPC Service Controls, CMEK, data residency, and per-model IAM consent gating.
A real contractual baseline. Google's Service Specific Terms training restriction: customer data is not used to train or fine-tune models without prior permission, GA and pre-GA alike.
Confidential-GPU infrastructure for self-deployed models. Confidential Space with NVIDIA H100 GPUs went GA and Confidential G4 VMs with Blackwell GPUs entered preview in June 2026, plus open-source Prompt Encryption SDKs.

The retention model, line by line

This is the load-bearing difference. The data-governance doc describes a real, auditable path to zero data retention. It is a checklist, not a default:

In-memory prompt caching is on by default for Gemini (24-hour TTL, project-isolated); disabling it takes a per-project cacheConfig change.
If safety classifiers flag activity, prompts can be logged up to 90 days under abuse monitoring. Master Agreement customers are exempt by default; others must file an exception request.
Designated "Advanced AI" models log all prompts and responses for up to 30 days, and opt-out "may not be possible". For Claude Fable 5 and Mythos 5 on Vertex, prompts and responses are retained up to 30 days, and sharing with Anthropic is mandatory.
Grounding with Google Search stores prompt-derived queries up to 3 days and cannot be disabled; Maps grounding stores 30 days. The Interactions API defaults to store=true.

TrustedRouter's realtime default needs no checklist: ordinary synchronous and streaming prompt paths do not touch persistent storage. We keep operational metadata — request ids, model and provider, token counts, latency, cost, region, API-key hash — and nothing of the prompt or output. Batch is a separate opt-in mode with enclave-encrypted retention up to 30 days. Details on /privacy.

Verify vs trust

Google documents no customer-verifiable TEE or attestation for the managed Gemini endpoint; the guarantee there is contractual. Its confidential-computing work applies when you self-deploy on Confidential VMs or GKE.

TrustedRouter's gateway runs inside TEEs on three clouds — GCP Confidential Space, AWS Nitro Enclaves, and Azure Confidential Containers — with live attestation endpoints and a one-command verifier at trust.trustedrouter.com. The gateway and control plane are source-available (BUSL-1.1, converting to Apache-2.0 four years after each release), so every line that touches your prompt is public, and the attestation binds the running build to published source and release digests.

The boundary: our attestation covers our gateway, not the model providers behind it. Downstream handling is tracked as cited policy tiers per provider — Google Vertex itself sits in the default candidate order of our trustedrouter/zdr route — and only trustedrouter/e2e restricts to providers (Tinfoil, Phala) running provider-side confidential compute with end-to-end encryption, where the provider's own enclave claims apply.

Pricing, with the bases stated

The two fee bases differ, so read carefully. Vertex bills the source token price, with real complexity: three service tiers, Global-versus-regional premiums, and promotional prices with expiry dates. Gemini 3.7/3.6 Flash's intro $0.75/$3.75 per 1M doubles January 1, 2027, and Claude Sonnet 5's $2.00/$10.00 promo ends August 31, 2026.

TrustedRouter prepaid bills the provider's token price plus 5.5%, with a $0.01/M floor; video is the provider quote plus 20%; there are no seat or subscription fees. So Gemini through us costs 5.5% more than direct Vertex — Gemini 2.5 Pro input at about $1.32/M instead of $1.25/M — and the fee buys cross-provider failover, privacy-tier routing, and the attested prompt path. BYOK is supported; our pricing page does not publish a separate BYOK fee. Exact per-route prices: /models and /pricing.

What we do not claim

Our attestation stops at the gateway. We do not attest Google's, Anthropic's, or anyone else's model servers.
We are young. Our public repos date from late April 2026 and public benchmark history starts June 2026; Google's operating record is measured in decades.
We publish a DPA, BAA, and subprocessor list, but no SOC 2 or HIPAA certification today. If FedRAMP High or certified HIPAA scope is a hard requirement, Vertex clears bars we do not.
We have no equivalent of Search or Maps grounding, Provisioned Throughput commitments, context-cache storage pricing, or an on-prem install.

Migration reality

Vertex exposes an OpenAI-compatible Chat Completions surface, but only for Gemini and select self-deployed Model Garden models; partner models such as Claude use provider-native APIs there, so those call sites need request rewrites either way. Its documented OpenAI-client auth uses one-hour Google OAuth access tokens refreshed programmatically; moving to TrustedRouter means a static key and one base URL, and adds Anthropic-style Messages alongside Chat Completions and Responses.

Two cautions from Google's docs: unsupported parameters on the Vertex OpenAI surface are dropped without an error, so a passed setting may never have been active, and Gemini-specific features that ride in extra_body need re-testing on any gateway. We publish a step-by-step migration doc only for OpenRouter today; coming from Vertex, start at /docs and the per-route prices on /models.

Migration shape

Start with one real request.

Vertex's OpenAI-compatible endpoint covers Gemini and select self-deployed models only, and its documented auth uses one-hour Google OAuth tokens; moving means a static TrustedRouter key, one base URL, and model-id renames. Partner-model call sites (Claude on Vertex) use provider-native APIs and need request-shape rewrites either way. Grounding and Provisioned Throughput have no TrustedRouter equivalent; Vertex's stored context caches map only to pass-through provider prompt caching on our side.

Keep the first test small, stream the response, and compare output, latency, provider selection, and billed usage before moving production traffic.

Agent setup Run a small eval

TrustedRouter sideOpenAI SDK

from openai import OpenAI

client = OpenAI(
    base_url="https://api.trustedrouter.com/v1",
    api_key="sk-tr-v1-...",
)

with client.chat.completions.create(
    model="trustedrouter/zdr",
    messages=[{"role": "user", "content": "Reply PONG"}],
    stream=True,
) as response:
    for chunk in response:
        print(chunk.choices[0].delta.content or "", end="")

Official evidence

Sources checked August 16, 2026

Report a change

Gemini data governance and zero-data-retention checklisthttps://docs.cloud.google.com/vertex-ai/generative-ai/docs/data-governance
Vertex abuse-monitoring prompt logging (up to 90 days)https://docs.cloud.google.com/vertex-ai/generative-ai/docs/learn/abuse-monitoring
Vertex generative AI pricinghttps://cloud.google.com/vertex-ai/generative-ai/pricing
OpenAI-compatible Chat Completions for Geminihttps://docs.cloud.google.com/vertex-ai/generative-ai/docs/multimodal/call-gemini-using-openai-library
Vertex AI to Agent Platform name changeshttps://docs.cloud.google.com/gemini-enterprise-agent-platform/vertex-ai-name-changes
Model Garden catalog (200+ models)https://cloud.google.com/model-garden
Google confidential computing announcements, June 2026https://cloud.google.com/blog/products/identity-security/verifiable-trust-in-the-ai-era-whats-new-in-confidential-computing
Google Cloud HIPAA-included productshttps://cloud.google.com/security/compliance/hipaa
TrustedRouter monthly benchmark reportsStable first-party route evidence
TrustedRouter live trust recordRelease and attestation evidence

All gateways

TrustedRouter vs Amazon BedrockCloud model platform
TrustedRouter vs Microsoft Azure AI FoundryCloud model platform
TrustedRouter vs OpenRouterHosted model marketplace
TrustedRouter vs Vercel AI GatewayHosted AI gateway

Questions

Is Gemini cheaper on Vertex AI or through TrustedRouter?

Vertex, by exactly our fee. TrustedRouter prepaid bills the provider's token price plus 5.5% with a $0.01/M floor, so Gemini 2.5 Pro input (up to 200K-token context) is $1.25/M direct on Vertex and about $1.32/M through us. Vertex's batch discount (~50%), context-cache storage pricing, and Provisioned Throughput commit pricing have no TrustedRouter equivalent — if you run only Gemini at scale, Vertex wins on price. BYOK is supported on our side; the pricing page does not publish a separate BYOK fee.

Can Vertex AI actually do zero data retention?

Yes, with work. Google documents the checklist: disable the default 24-hour in-memory cache per project, request an abuse-logging exception (Master Agreement customers are exempt by default), set store=false on the Interactions API, and avoid Search/Maps grounding retention. It may still be unavailable on some Advanced AI models — Claude Fable 5 and Mythos 5 on Vertex carry prompt and response retention of up to 30 days plus mandatory sharing with Anthropic. TrustedRouter's realtime default is no durable prompt or output logs, with only operational metadata retained.

Does TrustedRouter's attestation cover the model answering my prompt?

No. Attestation proves the gateway at api.trustedrouter.com runs the published source-available build; downstream providers are covered by cited retention-policy tiers, and only the trustedrouter/e2e route restricts to providers running their own confidential compute with end-to-end encryption (Tinfoil, Phala). Vertex's managed Gemini endpoint documents no attestation option; Google's confidential-GPU infrastructure applies when you self-deploy models on Confidential VMs or GKE.