What happened

OpenAI disclosed plans to make its most advanced large-language models available on Google Cloud. The models will be listed in Alphabet’s Model Garden, a catalogue that hosts more than 200 neural networks from Google, partners and the open-source community. The first model to appear is GPT-6 Astra, the company’s flagship LLM that has driven recent AI-generated math discoveries and achieved a perfect score on the ExploitBench security benchmark. OpenAI also added two smaller variants, GPT-6.1 Sol and GPT-6 Luna, which were launched in late September. Sol and Luna are priced five and one hundred times lower than Astra, respectively, offering a cheaper token cost at the expense of some output quality.

Why it matters

Astra includes safety optimisations that OpenAI says lead to “fewer misaligned outcomes” than other frontier models it has evaluated. The model is also configured to respect user-defined task restrictions, making it more suitable for enterprise deployments. By using Google Cloud’s Model Garden, customers can bundle the model fees with existing long-term spend commitments, simplifying procurement and avoiding separate AI budgets. The models will later be integrated into Google’s Confidential Compute services, which protect workloads with trusted execution environments (TEE) that encrypt memory segments for specific workloads. This adds a layer of security for organisations that need to protect sensitive data while training or running AI models.

What this means for enterprise IT teams

Enterprise IT teams gain a broader selection of high-performance LLMs without having to manage separate infrastructure. The pricing tiers for Sol and Luna allow teams to experiment with AI at lower cost, while Astra remains available for workloads that demand the highest quality and security. Integration with Google’s Antigravity coding tool and Gemini Agent means the models can be embedded into software development pipelines and business-process automation, reducing the effort required to build custom AI solutions. The use of Confidential Compute also eases compliance concerns when sharing datasets across organisational boundaries, offering strong data protection.

The bottom line

OpenAI’s partnership with Google Cloud opens its most advanced GPT-6 models to a wider enterprise audience, combining strong AI performance with built-in safety and security features. The tiered pricing and integration with Google’s existing cloud services give IT teams flexibility and cost control, while the planned move to Confidential Compute adds strong data protection. Enterprises can now choose the model that best fits each workload, from high-value, secure tasks on Astra to cost-effective experiments on Sol and Luna.

Sources

  • siliconangle.com