OpenAI Brings GPT-6 Astra and Scaled Models to Google Cloud
OpenAI plans to launch GPT-6 Astra, Sol, and Luna on Google Cloud's Model Garden, integrating them with Confidential Compute and developer tools.
OpenAI Group PBC is set to introduce its latest generation of large language models to Google Cloud, marking a notable expansion across enterprise cloud ecosystems. Through this partnership, the artificial intelligence research lab will offer its flagship systems via Model Garden, Google Cloud’s repository that currently hosts more than 200 neural networks spanning Google's proprietary tools, open-source options, and specialized task models like translation networks.
Leading the rollout is OpenAI’s premier model, GPT-6 Astra. The model has recently been highlighted by the company for generating several AI-driven mathematical breakthroughs. In addition to mathematical problem solving, Astra registered a perfect score on ExploitBench, an industry benchmark designed to assess how effectively language models identify and leverage cybersecurity vulnerabilities. To address corporate governance concerns, the system incorporates enterprise-focused safety adjustments; OpenAI reports that Astra generates fewer misaligned outcomes relative to all other frontier models it tested, and it is explicitly structured to respect user-imposed operating boundaries without bypassing task restrictions.
Tiered model options and pricing
Alongside Astra, Google Cloud users will gain access to two smaller variants released in late September: GPT-6.1 Sol and GPT-6 Luna. Both models provide developers with lower-cost alternatives by accepting modest reductions in generation quality. In terms of token expense, Sol is priced five times lower than Astra, while Luna reduces the pricing barrier further, costing 100 times less than the flagship.
Enterprise billing processes will also be streamlined. Google Cloud clients will be able to apply existing long-term spending commitments toward the usage of OpenAI’s models, eliminating the friction of negotiating separate vendor contracts or approving distinct AI budgets. Explaining the motivation behind the integration, Google Cloud Chief Executive Thomas Kurian noted that the provider is focused on offering customers an expansive array of models, including systems from OpenAI, allowing teams to balance quality against expenses for specific workloads.
Infrastructure security and tooling integrations
Following the rollout across Model Garden over the coming weeks, the two companies plan to incorporate Astra, Sol, and Luna into Google’s Confidential Compute infrastructure. This environment relies on server-grade CPU security features known as Trusted Execution Environments (TEEs), which isolate and encrypt specific memory segments on the host server so only designated workloads can access them. This configuration offers enhanced protection for proprietary corporate workloads and facilitates secure collaborations, such as allowing multiple organizations to share sensitive data for AI training tasks.
Beyond foundational infrastructure, the models are scheduled to interface directly with Google’s software applications. These include Antigravity, a vibe coding product that leverages ensembles of autonomous agents to accelerate programming workflows, and Gemini Agent, a solution tailored to automate operational routines that traverse multiple cloud programs.
What it means for developers
The arrival of OpenAI's models inside Google Cloud offers engineering teams substantial architectural flexibility. Rather than managing disparate accounts across different cloud ecosystems, developers can orchestrate GPT-6 Astra alongside Google Cloud services while drawing against their existing committed cloud expenditures.
Furthermore, the tiered pricing structure between Astra, Sol, and Luna gives engineers practical architectural levers. High-stakes tasks—such as cybersecurity audits or advanced analytical reasoning—can be routed to Astra, while cost-sensitive, high-volume production jobs can be delegated to Sol or Luna. For teams that want to test and benchmark various options across providers without dealing with multiple vendor silos, developers can try top AI models cheaply through one API at https://apixoai.online.
With broader access to TEE-backed confidential computing, developers building in tightly regulated environments will also have an easier route to deploying frontier models without exposing sensitive memory contents or complicating external data partnerships.
Source: OpenAI to make its most advanced AI models available in Google Cloud — SiliconANGLE AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
