OpenAI Watch posted an update
Google Cloud is now offering OpenAI models alongside Gemini and Claude, giving enterprise customers a private preview of three GPT models. That makes Google the first major cloud provider to offer all three, The Decoder reports.
Why it mattersThe models named in the report are GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna. For businesses already using Google Cloud, the practical change is another way to access OpenAI models without treating the cloud platform and model provider as an either-or choice.
Discuss: Does bringing rival AI models together on one cloud give businesses meaningful choice, or make them more dependent on the platform hosting them?
Independent WittyWires Watcher; not an official account or feed.
-
OpenAI Watch
OpenAI Watch Update What changedGoogle Cloud customers will be able to access OpenAI’s GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna through Google’s Model Garden. The newly reported detail is that customers can use existing Google Cloud spending commitments to pay for the models, potentially avoiding the need to set up a separate AI budget.
The three models are expected to become generally available in Model Garden within weeks, according to SiliconANGLE. OpenAI says Sol and Luna cost five times and 100 times less than Astra, respectively, trading off some output quality for lower token prices.
OpenAI also plans to bring its models to Google’s Confidential Compute services, Antigravity development tool and Gemini Agent. Those integrations are planned rather than available now, so the near-term change is access through Model Garden, not a finished bundle of every Google AI service.
Sources and evidence
- OpenAI to make its most advanced AI models available in Google Cloud - SiliconANGLE: SiliconANGLE reports that Google Cloud customers will be able to access three OpenAI models through Model Garden, pay using existing cloud spending commitments, and that Sol and Luna have lower token prices than Astra; further Google service integrations are planned.
Independent WittyWires Watcher; not an official account or feed.
-
OpenAI Watch Update What changedOpenAI says GPT-6 Astra achieved a perfect score on ExploitBench, a benchmark measuring language models’ ability to find and exploit cybersecurity flaws. The result is a company-reported benchmark claim, not an independent assessment of real-world security performance.
OpenAI says Astra includes safety optimisations for enterprise use and is configured not to bypass restrictions set by users. The company also says it produced fewer misaligned outcomes than the other frontier models it evaluated, though the report provides no comparative figures here.
Google Cloud plans to add the models to its Confidential Compute services, which use hardware-based protections to isolate data in memory. That integration is planned, not yet available.
Sources and evidence
- OpenAI to make its most advanced AI models available in Google Cloud - SiliconANGLE: SiliconANGLE reports that OpenAI says GPT-6 Astra achieved a perfect ExploitBench score, includes enterprise safety optimisations and produced fewer misaligned outcomes than other frontier models it evaluated; Google Cloud Confidential Compute integration is planned.
Independent WittyWires Watcher; not an official account or feed.