Google Cloud announced on October 21, 2026, that GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna were rolling out in private preview for selected enterprise customers through Model Garden. The invite-only program lets eligible customers evaluate the models for supported development and internal testing, while also offering a route for certain Codex workflows to use Google Cloud for inference.
Who can evaluate the models
OpenAI describes the preview as invite-only. Selected customers may evaluate the models for supported development and internal-testing use cases, according to the OpenAI preview-access form. The form also lists Codex access through the API, Codex app, CLI and supported IDEs.
The preview is a limited evaluation program for selected enterprise customers. OpenAI said on October 21 that general availability was planned for the coming weeks.
How Google Cloud describes the three models
Google Cloud positions GPT-6 Astra for demanding coding, computer use, professional work and scientific research. It describes GPT-6.1 Sol as offering near-Astra intelligence at lower cost for agentic coding, computer use and professional work. GPT-6 Luna is described as fast and cost-efficient for everyday work at scale.
These descriptions give teams a starting point for matching a model to a task: Astra for more demanding work, Sol for specified agentic and professional tasks at lower cost, and Luna for everyday workloads at scale. They are Google Cloud’s descriptions of the models, not comparative benchmark results.
For background on the model itself, read our earlier coverage of GPT-6 Astra’s computer-use launch. The Google Cloud announcement adds an enterprise distribution route for Astra and the other named models.
Codex routing and Marketplace billing
Google Cloud says local Codex CLI and Codex Desktop workflows can connect to its Gemini platform for inference. Inference is the process of running a model to generate a response. That route lets a team work in those local Codex tools while sending inference through Google Cloud.
Purchases go through Google Cloud Marketplace and appear on one Google Cloud invoice. Google Cloud says the spending draws down existing cloud commitments and counts toward Flexible Savings Plans. The announcement does not say that those plans reduce the models’ rates.
Data controls and OpenAI’s cloud relationship
Google Cloud says customer prompts, context and outputs stay within the designated data perimeter and are not used to train foundation models. It also names hardened isolated gateways, authenticated identities, VPC Service Controls and unified audit logging among its controls. Google Cloud described Zero Data Retention options and Confidential Compute as coming soon.
OpenAI said Microsoft remains its primary cloud partner. The Google Cloud preview adds another route for selected customers to access OpenAI models and use the specified Codex workflows.
