OpenAI's GPT-6 Astra to land on Google Cloud Model Garden

The arrangement pairs two companies that compete at the model layer but can cooperate at the distribution layer. For Google, adding OpenAI's models to Model Garden addresses enterprise customers who want to use several AI models side by side rather than commit to one vendor. For OpenAI, the deal is a way to expand adoption of its models among companies already running on Google Cloud.

The three models are positioned for different workloads: GPT-6 Astra for coding and advanced scientific research, GPT-6.1 Sol for specialized work with an emphasis on cost performance, and GPT-6 Luna as a low-cost option for everyday tasks. They are slated to work in Google Antigravity and Gemini Agent, and can be reached through Google's API or by connecting local Codex CLI and desktop workflows directly to the Gemini platform for inference.

Google is also emphasizing the robustness of the Gemini platform, saying it strictly manages data and access boundaries and provides high visibility and transparency into how AI agents and models behave. It guarantees that customer prompts, context and outputs will not be used to train the underlying models — a point likely to matter to enterprise buyers weighing whether to route sensitive work through a third-party model inside Google's environment. General enterprise availability is expected within the next few weeks, with applications handled through a dedicated form from both companies.