Zero data retention AI: which models run on the Private Lane

What zero data retention means, and what it does not

When you send a prompt, the company that runs the model has to process it to answer. Retention is about what happens next. Under a standard agreement, a provider may keep prompts and outputs for a period, often for abuse monitoring. Under a Zero Data Retention agreement, it does not store them after the response. Private Lane chats are sent with Zero Data Retention, and only to providers covered by a ZDR agreement. Standard Lane models follow that provider's standard retention terms. Quad asks you to confirm before your first Standard Lane message, so you never move lanes by accident. ZDR is about the model provider. Quad itself still stores your chat history so you can come back to it, encrypted in transit and at rest, unless you use an Incognito chat. Quad never uses your conversations or files to train its own models, on either lane.

Private Lane and Standard Lane by lab

Generated from the catalog. 14 of 18 labs run every model they have in Quad on the Private Lane. OpenAI (Private Lane (Zero Data Retention): GPT-6.1 Sol, GPT-6 Astra, GPT-6 Sol, GPT-6 Luna, GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna, GPT-5.5 Thinking, GPT-5.5 Pro, GPT-5.3 Codex, gpt-oss 120B, gpt-oss 20B; Standard Lane: None). Google (Private Lane (Zero Data Retention): Gemini 4 Argon, Gemini 3.8 Flash, Gemini 3.1 Pro, Gemini 3.1 Flash Lite; Standard Lane: None). Anthropic (Private Lane (Zero Data Retention): Claude Haiku 5.5, Claude Opus 5.5, Claude Sonnet 5.5; Standard Lane: Claude Fable 5). Moonshot AI (Private Lane (Zero Data Retention): Kimi K3, Kimi K2.7 Code; Standard Lane: None). Meta (Private Lane (Zero Data Retention): Muse Wonder, Muse Spark 1.3, Muse Glimmer 30B; Standard Lane: None). Thinking Machines (Private Lane (Zero Data Retention): Inkling, Inkling Small; Standard Lane: None). Z.ai (Private Lane (Zero Data Retention): GLM 5.3, GLM 5.3 Flash; Standard Lane: GLM 5.3 FlashX). xAI (Private Lane (Zero Data Retention): Grok 4.7, Grok 4.1 Fast, Grok 4.1 Fast Reasoning; Standard Lane: None). MiniMax (Private Lane (Zero Data Retention): MiniMax M3; Standard Lane: None). Mistral AI (Private Lane (Zero Data Retention): Mistral Medium 3.5, Mistral Large 4 Preview; Standard Lane: None). Amazon (Private Lane (Zero Data Retention): Nova 2 Lite; Standard Lane: None). NVIDIA (Private Lane (Zero Data Retention): Nemotron 3.5 Lightning 30B, Nemotron 3 Ultra; Standard Lane: None). Xiaomi (Private Lane (Zero Data Retention): MiMo V2.6 Pro, MiMo V2.6 Flash; Standard Lane: None). Inception (Private Lane (Zero Data Retention): None; Standard Lane: Mercury 2.5). Meituan (Private Lane (Zero Data Retention): None; Standard Lane: LongCat 2.5 Preview). Fireworks (Private Lane (Zero Data Retention): Ember 1; Standard Lane: None). DeepSeek (Private Lane (Zero Data Retention): DeepSeek V4.1 Flash, DeepSeek V4 Pro, DeepSeek V4 Flash; Standard Lane: None). Alibaba (Private Lane (Zero Data Retention): Qwen 3.8 Max, Qwen 3.8 Flash, Qwen 3.8 27B, Qwen 3.8 Omni Flash; Standard Lane: None).

The models that are Standard Lane only

Claude Fable 5 (Lab: Anthropic; Available on: Ultra plan and up). GLM 5.3 FlashX (Lab: Z.ai; Available on: Free plan and up). Mercury 2.5 (Lab: Inception; Available on: Free plan and up). LongCat 2.5 Preview (Lab: Meituan; Available on: Free plan and up). These 4 models are available only through routes without a ZDR agreement. They still work in Quad, and some are worth using for non-sensitive work. If a conversation holds anything confidential, pick a Private Lane model instead, and decline the confirmation if it appears. Image and video models in Studio also run on the Standard Lane, so the provider may keep prompts, reference images and results under its own policy. Studio asks you to confirm this once before your first generation, and chat keeps the lane of the model you picked.

Zero retention on a budget

28 of the Private Lane models are on the Free plan, so you can test zero-retention chat without paying. Among them are GPT-6 Luna, Claude Haiku 5.5, Gemini 3.1 Flash Lite, DeepSeek V4 and Kimi K2.7 Code. The frontier models such as Claude Opus 5.5 and GPT-6.1 Sol are on the Private Lane too, from the plan each one needs. Lanes can change when a provider signs or ends a ZDR agreement. Each model page shows the current lane, and this page regenerates from the same catalog.

How to keep a sensitive conversation private

Choose a model from the Private Lane column above. The model picker and each model page show the lane. Decline the Standard Lane confirmation if it appears in a sensitive chat, and switch models instead. Use an Incognito chat when you do not want the conversation saved to your history. New uploads are off in Incognito. Export or delete chats yourself at any time from your account. Remove names, account numbers and other identifiers you do not need. No retention policy protects data you did not have to send.

Frequently asked questions

What is zero data retention in AI?

It is an agreement under which the model provider does not store your prompts or the model's outputs after it responds. It does not mean the provider never sees the prompt, since it has to process it to answer.

Which AI models have zero data retention?

In Quad, 47 of 51 chat models run on the Private Lane with Zero Data Retention, including GPT, Claude, Gemini, Grok, DeepSeek and Kimi models. The table on this page lists every one by lab.

Does Claude have zero data retention?

Anthropic offers ZDR to some API customers. In Quad, Claude Haiku 5.5, Claude Opus 5.5, Claude Sonnet 5.5 run on the Private Lane. Anthropic models not listed there use the Standard Lane.

Does Quad store my chats if the model has zero retention?

Yes, unless you use Incognito. ZDR covers the model provider. Quad keeps your chat history so you can return to it, encrypted in transit and at rest, and you can export or delete it whenever you like.

What happens when I pick a Standard Lane model?

Quad asks you to confirm before your first message. After that, the provider's standard retention terms apply to that conversation. Quad still does not train on it.