OpenAI to make its most advanced AI models available in Google Cloud

OpenAI is bringing its latest large language models to Google Cloud, starting with Model Garden, Alphabet’s catalog of more than 200 neural networks from Google, its partners and the open-source world. The first model in line is GPT-6 Astra, OpenAI’s flagship system and the one behind several math discoveries the company has been talking about in recent weeks.

Astra also turned in a perfect score on ExploitBench, a test of how well a model can find and exploit security flaws. OpenAI says the model has been tuned for enterprise use, with safety changes meant to produce fewer misaligned outcomes than the other frontier models it has evaluated. It is also set up to respect user-defined limits instead of sidestepping them.

Google Cloud customers won’t just get Astra. OpenAI is also offering GPT-6.1 Sol and GPT-6 Luna, smaller versions launched in late September that give up some quality in exchange for much lower token pricing. Sol costs five times less than Astra, and Luna costs 100 times less.

That pricing matters because Google Cloud buyers can use existing long-term spending commitments to pay for access to OpenAI’s models. In other words, they do not need to carve out a separate AI budget just to try them. Google Cloud chief executive Thomas Kurian framed the move as part of a broader push to give customers a wide choice of models for different jobs, with better quality where needed and lower costs where possible.

The rollout is not stopping at Model Garden. OpenAI and Google plan to bring the models into Google’s Confidential Compute services, which use trusted execution environments to protect workloads and can help when multiple organizations need to share data for AI training. OpenAI also plans to make the models available through Google’s Antigravity and Gemini Agent products, both aimed at automating software and business work across cloud apps.