OpenAI's Latest Models Coming to Google Cloud Through Multiple Services

OpenAI Group PBC revealed its intention to deploy its newest large language models across Google Cloud infrastructure. The models will be offered via Alphabet Inc.'s Model Garden service, which currently hosts more than 200 neural networks spanning Google's own offerings, partner technologies and open-source projects. Beyond large language models, the catalog encompasses specialized smaller models designed for specific applications including translation tasks.

The initial model OpenAI is introducing to Google Cloud is GPT-6 Astra, its flagship offering. This algorithm has driven several AI-generated mathematical breakthroughs that OpenAI has disclosed in recent weeks. The model also achieved a perfect score on ExploitBench, a testing framework that evaluates how effectively language models can identify and leverage cybersecurity vulnerabilities.

Astra incorporates safety mechanisms intended to make it suitable for business deployments. According to OpenAI, the model produces "fewer misaligned outcomes" compared to other frontier models the company has tested. The system is designed to respect user-imposed constraints on which tasks it will execute.

Customers on Google Cloud will gain access to two additional models: GPT-6.1 Sol and GPT-6 Luna. These are compact versions of Astra that OpenAI released in late September. Both sacrifice some response quality in exchange for substantially lower token costs. Sol and Luna are priced at five times and 100 times less than Astra, respectively.

We are focused on giving customers a broad selection of models, including those from OpenAI, so they can choose the right model for any given job, helping to deliver optimal quality and lower costs.

Google Cloud Chief Executive Thomas Kurian

Cloud service providers typically offer reduced rates to organizations committing to extended spending arrangements. OpenAI notes that Google Cloud customers can apply these commitments toward paying for its model access. This approach eliminates the requirement to establish a separate budget for artificial intelligence, streamlining the purchasing process.

Astra, Sol and Luna will reach general availability within Model Garden in the coming weeks. The two companies intend to subsequently integrate these models into Google's Confidential Compute portfolio of infrastructure offerings.

Enterprise-grade processors like those deployed by Google include a security mechanism known as TEE. This technology encrypts a portion of server memory in a way that restricts access to a particular workload only. Google's Confidential Compute services leverage this technology to protect customer operations. The services also facilitate scenarios such as enabling multiple organizations to share datasets for collaborative AI model training.

OpenAI intends to distribute its models through additional Google offerings as well. Antigravity, a code generation tool powered by agent ensembles, accelerates software development initiatives. Gemini Agent is built to handle automated business processes that operate across multiple cloud-based applications.