Hi Palantir team!
Quick follow-up to my July thread about GPT-5.6 and Grok 4.5 here. Both showed up in the docs the next day, so thanks for that.
Two more models have shipped since then, and I didn’t see either one in the September 10 batch that added Gemini 3.8 Flash, the Fireworks open-weight models, and GPT-5.6 for IL2, IL4, and IL5:
- GPT-6 Astra, announced September 3 and available in the OpenAI API as gpt-6-astra since September 4. It is also generally available in Microsoft Foundry (September 3) and on Amazon Bedrock (September 8).
- Claude Fable 5.1, released September 1 as claude-fable-5-1 on the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry.
A few questions:
- Is GPT-6 Astra queued up for enrollments with OpenAI Direct and/or Azure OpenAI enabled? If something is on the critical path, like regional availability or the newer Responses API parameters, that’s useful to know so we can plan around it.
- On Fable 5.1 and data retention: the reply in the July thread noted that every model in AIP runs with Zero Data Retention, which I took as the reason Fable 5 never showed up (it’s a Covered Model with 30-day retention and no ZDR option). Fable 5.1 looks different. Anthropic says eligible customers can use it with zero data retention until Enterprise Frontier Safeguards rolls out in phases later this fall, and EFS keeps the monitoring data in the customer’s own cloud infrastructure. Does that change things for AIP? I’m mainly wondering whether Palantir is eligible for ZDR on Fable 5.1 through the Anthropic, Bedrock, Vertex, and Azure channels AIP already uses, and whether EFS is something you’d pick up once it’s available.
- If either model is coming, does it land everywhere at once (AIP Logic, Functions and transforms, agents, Evals, the Use LLM node in Pipeline Builder, LLM proxy) or in stages?
We’d like to put both up against what we’re using today in AIP Evals and the new Pipeline Builder evaluation suites.
Thanks!