Today in DeAI: Anthropic launches Claude Fable 5.1 with an enterprise tier it says matches zero data retention, OpenRouter adds per-request ZDR enforcement, and DeepSeek's V4 Pro stays online past its planned shutdown date.
Anthropic ships Fable 5.1, promises ZDR-equivalent enterprise privacy
Anthropic released Claude Fable 5.1 this morning — its new flagship for coding and knowledge work — alongside Mythos 5.1, the same weights gated to vetted cybersecurity and life-science users. Cache reads drop 75% to $0.25 per million tokens, and the new Enterprise Frontier Safeguards store customer data in customer-controlled infrastructure, which Anthropic calls "complete privacy (the same as a zero data retention policy)" — a provider claim, with phased rollout starting this fall. Why it matters: an Anthropic-grade ZDR-equivalent would reshuffle the private-inference shortlist for regulated workloads, but the guarantee is unverified until the EFS terms and any audit commitments are published. (Anthropic) — our coverage
Anthropic's consumer terms split retention by consent
Days before the Fable launch, Anthropic updated its consumer terms: Free, Pro, and Max users who opt in to model improvement can have conversations retained up to five years, while opted-out users stay on a 30-day window. Commercial Terms customers (Team, Enterprise, API) remain no-training by default. Why it matters: Anthropic's enterprise and consumer privacy postures are moving in opposite directions in the same week — enterprise toward ZDR-equivalence, consumer toward longer retention — and the split is the fact procurement teams will ask about. (Anthropic)
OpenRouter adds per-request zero-retention routing
OpenRouter now documents a zdr boolean in the provider request object that restricts routing to endpoints honoring zero data retention, alongside an account-level setting. The documented boundary is the story: ZDR governs provider-side retention only and does not control OpenRouter's own router-side logging or caching. Why it matters: builders can now flip a per-request privacy control at the router layer, but "ZDR via OpenRouter" is not equivalent to a private endpoint — the guarantee stops at the provider, per OpenRouter's own docs. (OpenRouter)
OpenAI Agents API beta reportedly keeps session state on OpenAI infrastructure
OpenAI's new Agents API public beta ships with zero data retention unsupported, US-only data residency, and session state retained on OpenAI infrastructure even when execution runs in a self-hosted sandbox, per secondary reporting at Yusmp Group. Why it matters: if confirmed against OpenAI's docs, that is the exact gap that decides whether agent workloads can leave regulated environments — the counterexample to OpenAI's frontier-model ZDR reaffirmation. Treat as a claim until the docs are quoted.
Iris open-weight search agents lead their size class
The AllSpark team released Iris-mini and Iris-pro, open-weight search agents built on Qwen models that lead benchmarks among open-weight models in their size classes, with reported transfer gains on untrained tool-use and office tasks, per The Decoder. Why it matters: builders assembling self-hosted agent stacks currently rent closed search APIs; a competitive open-weight option on Qwen licensing terms is directly relevant. The benchmark leads are the authors' claims until independent runs land.
Watching tomorrow
Whether Anthropic publishes EFS terms with any audit or abuse-logging detail as the phased enterprise rollout begins — that documentation is what turns today's ZDR-equivalence claim into something checkable.
Sources
- Claude Fable 5.1 and Mythos 5.1 — Anthropic
- Updates to our consumer terms — Anthropic
- Provider Routing — OpenRouter Docs — OpenRouter
- OpenAI Agents API public beta — Yusmp Group
- Iris-mini and Iris-pro are the strongest open-weight search agents in their class — The Decoder
- DeepSeek API Models & Pricing — DeepSeek
About DeAI
DeAI is an independent publication covering open-weight AI models, private inference, and decentralized infrastructure — the tools for running AI you actually control. We test providers on price, privacy, and refusal behavior and publish the numbers, not the vibes. DeAI is powered by Morpheus (mor.org), a decentralized inference marketplace, and covers it on the same terms as every other provider.
Powered by Morpheus and StrandCMS
Morpheus is a decentralized inference marketplace, covered on the same terms as every other provider — we rank it wherever the data lands. StrandCMS is the open-source, agent-first framework this site is built on.
