---
title: "LLM API data retention and training policies compared"
description: "What OpenAI, Anthropic, Gemini API, Mistral and BroadNet officially say about storing API prompts and training on them, quoted from their pages in November 2026."
url: https://broadrouter.com/blog/llm-api-data-retention-compared/
lang: en
updated: 2026-11-14
publisher: BroadRouter (a BroadNet site); API by BroadNet Inference, https://inference.broadnet.ai
---

# LLM API data retention compared: OpenAI, Anthropic, Gemini, Mistral

By the BroadNet Inference team · 6 min read · Published 2026-11-14 · Updated 2026-11-15

As of 14 November 2026, OpenAI keeps API abuse-monitoring logs for up to 30 days and Anthropic deletes API content within 30 days; neither trains on it by default. Mistral keeps it 30 rolling days for abuse monitoring. Google logs paid Gemini API prompts for "a limited period" without training on them. BroadNet Inference says it keeps no content and does not train.

## Key takeaways

-   Default retention is not zero at most large providers: 30 days is common, for abuse monitoring.
-   Zero data retention exists at OpenAI, Anthropic and Mistral, but it needs approval or a sales contract and covers only some endpoints.
-   Free tiers differ: Google may use unpaid Gemini API content to improve products, and Mistral may train on free-mode data unless you opt out.
-   Stateful features (files, batches, conversations, agents) keep data until you delete it, even under ZDR.
-   BroadNet Inference says it keeps no prompts or outputs and does not train, fine-tune or evaluate on inference data.

**On this page**

1.  How do OpenAI, Anthropic, Gemini and Mistral compare?
2.  How long does OpenAI keep API data?
3.  How long does Anthropic keep API data?
4.  Does Google keep or train on Gemini API prompts?
5.  How long does Mistral keep API data?
6.  What does BroadNet Inference keep?
7.  How should you read these policies?
8.  Where BroadNet Inference fits

As of 14 November 2026, OpenAI keeps API abuse-monitoring logs up to 30 days, Anthropic deletes API content within 30 days, Mistral keeps it 30 rolling days, and Google logs paid Gemini prompts for “a limited period”. OpenAI, Anthropic and paid Gemini do not train on it by default; Mistral’s paid default is unstated. BroadNet Inference keeps no content and does not train.

This post compares what each provider’s official pages say. If you want the questions to put to any provider, use the [zero data retention checklist](https://broadrouter.com/blog/zero-data-retention-llm-api-checklist/); for the contract side, the [GDPR checklist](https://broadrouter.com/blog/gdpr-llm-api-checklist/).

We fetched each provider’s official page on **14 November 2026** and quoted it directly. Where a page does not say something, we say so instead of guessing. Re-check before you sign anything: these pages change.

## How do OpenAI, Anthropic, Gemini and Mistral compare?

| Provider (API) | Default retention of content | Training on API content | Zero data retention |
| --- | --- | --- | --- |
| OpenAI | Abuse monitoring logs “retained for up to 30 days” | Not used “unless you explicitly opt in” | Eligible endpoints only, “subject to prior approval” |
| Anthropic | Deleted “within 30 days of receipt or generation”, with exceptions | “may not train models on Customer Content” (Commercial Terms) | Per organisation, via sales; not for Covered Models |
| Google Gemini API (paid) | Logged “for a limited period of time” for abuse | “Google doesn’t use your prompts … or responses to improve our products” | Not described on the terms page |
| Google Gemini API (unpaid) | Not specified | Used to “provide, improve, and develop” products; human review possible | Not described |
| Mistral | Kept for “thirty (30) rolling days to monitor abuse” | Free mode may be used; opt-out available | Paid plans, stateless calls, on approval |
| BroadNet Inference | Gateway “keeps no prompts, messages, images, audio, responses or transcripts” | “We do not use your inference data to train, fine-tune or evaluate models” | Default |

Our open dataset of [LLM API free tiers and rate limits](https://broadrouter.com/data/llm-api-free-tiers-and-rate-limits/) compares what 14 providers publish about free access, starting rate limits, card requirements and training on free-tier data, with the source and date for each value.

## How long does OpenAI keep API data?

OpenAI’s [data controls page](https://developers.openai.com/api/docs/guides/your-data) says data sent to the API “is not used to train or improve OpenAI models” unless you opt in. “By default, abuse monitoring logs are generated for all API feature usage and retained for up to 30 days,” and longer where the law requires it.

Some features store **application state** separately:

-   The Responses API has “a 30 day Application State retention period by default” when `store` is true.
-   Files, conversations and threads are kept until you delete them.
-   Assistants objects are deleted 30 days after you delete them.

[Zero data retention](https://broadrouter.com/glossary/zero-data-retention/) and Modified Abuse Monitoring “are subject to prior approval by OpenAI and acceptance of additional requirements”. Eligibility is per endpoint: Chat Completions and Responses are eligible, while Files, Batches and Fine-tuning are not.

## How long does Anthropic keep API data?

Anthropic’s [privacy center](https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-commercial-api-data) (updated July 1, 2026) says: “we automatically delete inputs and outputs on our backend within 30 days of receipt or generation”, except for services with longer retention under your control (such as the Files API), a zero data retention agreement, Usage Policy enforcement or the law. Flagged content may be kept “for up to 2 years”.

On training, the [Commercial Terms](https://www.anthropic.com/legal/commercial-terms) say: “Anthropic may not train models on Customer Content from Services.”

[ZDR](https://platform.claude.com/docs/en/manage-claude/api-and-data-retention) is requested through Anthropic’s sales team and enabled per organisation. One newer condition: some models, which Anthropic calls **Covered Models** (Claude Fable 5, 5.1 and 5.2 and Claude Mythos 5, 5.1 and 5.2, as listed on that page on 15 November 2026), “require 30-day data retention. ZDR is therefore not available for any of them unless expressly authorized by Anthropic.”

## Does Google keep or train on Gemini API prompts?

Google’s [Gemini API terms](https://aistudio.google.com/docs/terms) (effective March 23, 2026) split by tier:

-   **Paid services:** “Google doesn’t use your prompts” (including system instructions, cached content and files) “or responses to improve our products”, and “logs prompts and responses for a limited period of time, solely for detecting and preventing violations”. No number of days is given.
-   **Unpaid services:** Google uses content “to provide, improve, and develop” its products, and “human reviewers may read, annotate, and process your API input and output”. The terms say: “Do not submit sensitive, confidential, or personal information to the Unpaid Services.”
-   **EEA, UK and Switzerland:** the paid-services data terms apply to all services for users there.

Separately, the [logs policy](https://aistudio.google.com/docs/logs-policy) describes optional project logs you can view, kept for “a default maximum period of 55 days” and adjustable to 7, 14, 28 or 55 days. Those logs are not used for training unless you share a dataset with Google or leave feedback such as a thumbs up or down.

## How long does Mistral keep API data?

Mistral’s [privacy policy](https://legal.mistral.ai/terms/privacy-policy) (effective September 3, 2026) says that, except for specific APIs, it keeps API input and output “for the period necessary to generate the Output and then for thirty (30) rolling days to monitor abuse (unless zero data retention is activated)”. Agents API data is kept until you close your account, and fine-tuning data until you delete it.

On training, the [help center](https://help.mistral.ai/en/articles/347617-do-you-use-my-user-data-to-train-your-artificial-intelligence-models) says that in free mode “we may use your data (input and output)” to train models, with an opt-out. For pay-as-you-go it says customers “have the right to opt out at any time”; it does not state the default. Its [opt-out article](https://help.mistral.ai/en/articles/455207-can-i-opt-out-of-my-input-or-output-data-being-used-for-training) puts the API opt-out under Privacy, “Anonymous improvement data”, in the Admin panel.

Mistral’s [ZDR docs](https://docs.mistral.ai/admin/monitor-comply/zero-data-retention) say ZDR “is available on paid plans for supported stateless API calls”, that Mistral reviews each request, and that “ZDR and training opt-out are separate controls”.

## What does BroadNet Inference keep?

BroadNet’s [privacy statement](https://www.broadnet.ai/inference-privacy) (effective September 2026) says the gateway “keeps no prompts, messages, images, audio, responses or transcripts” and “We do not use your inference data to train, fine-tune or evaluate models.” It keeps content-free traffic and usage records “for as long as billing, tax, security and legal obligations require”. There is no approval step: it is the default for every key. The full list of kept fields is on our [zero data retention page](https://broadrouter.com/zero-data-retention/).

## How should you read these policies?

1.  **Separate retention from training.** A provider can promise no training and still keep content for 30 days.
2.  **Check stateful features.** Files, batches, stored responses and agents keep data until deleted, under every provider here.
3.  **Check your tier.** Free and paid terms differ at Google and Mistral.
4.  **Check the model.** Anthropic’s Covered Models rule shows retention can depend on the model you call.
5.  **Get it in a contract.** A help-center page can change. Your DPA and order form are what bind the provider. Use our [GDPR checklist](https://broadrouter.com/blog/gdpr-llm-api-checklist/) for the contract questions and the [ZDR checklist](https://broadrouter.com/blog/zero-data-retention-llm-api-checklist/) for the technical ones.

## Where BroadNet Inference fits

If your requirement is “no content stored, no training, without an approval process”, BroadNet’s default matches it, and its OpenAI-compatible API means a [base URL](https://broadrouter.com/glossary/base-url/) change for most apps. BroadNet runs [open-weight models](https://broadrouter.com/glossary/open-weight-model/), not GPT, Claude or Gemini, so if you need those specific models, you will be working through the policies above. If you are moving off OpenAI for privacy reasons, the [OpenAI API alternative page](https://broadrouter.com/alternatives/openai-api-alternative/) compares the options. For sensitive workloads on any provider, redact first: see [sending sensitive data to an LLM API](https://broadrouter.com/blog/llm-api-sensitive-data/).

Free tools for this

-   [OpenAI-compatible API request builder](https://broadrouter.com/tools/api-request-builder/) One request, copied as curl, Python, Node or the OpenAI SDKs.
-   [LLM API cost calculator](https://broadrouter.com/tools/llm-api-cost-calculator/) Price a workload on every model at once, with cache hits, reasoning and a flat-plan check.

[All free tools](https://broadrouter.com/tools/)

## Terms in this article

[Zero data retention](https://broadrouter.com/glossary/zero-data-retention/)

Zero data retention (ZDR) is a data-handling policy in which an AI provider processes prompts and outputs only to serve a request and does not store that content afterward, in logs, caches or training data. Content-free records such as token counts and timings are usually still kept.

[Fine-tuning](https://broadrouter.com/glossary/fine-tuning/)

Fine-tuning is the process of continuing to train a pre-trained language model on a smaller, task-specific dataset, such as example conversations, so that its weights change and it performs that task or style more reliably without needing long instructions or examples in every prompt.

[GDPR](https://broadrouter.com/glossary/gdpr/)

The GDPR (General Data Protection Regulation, Regulation (EU) 2016/679) is the EU law on processing personal data. It has applied since 25 May 2018 and covers EU-established organisations and some outside the EU. For LLM APIs it governs processor contracts, international transfers and data minimisation.

[Rate limit](https://broadrouter.com/glossary/rate-limit/)

A rate limit is a cap an API places on how much a client can use it in a time window, such as requests per minute or tokens per minute. Requests over the limit are refused, usually with HTTP 429 Too Many Requests, until usage falls back under it.

[Base URL](https://broadrouter.com/glossary/base-url/)

A base URL is the fixed root address of an API, such as https://api.example.com/v1, that a client puts in front of every endpoint path, such as /chat/completions. In OpenAI-compatible SDKs, changing the base URL, together with the API key and model ID, points the same code at a different provider.

[Chat Completions API](https://broadrouter.com/glossary/chat-completions-api/)

The Chat Completions API is OpenAI's HTTP endpoint, POST /v1/chat/completions, that takes a model name and a list of role-tagged messages and returns the model's next message. Its request and response shape became the common format that OpenAI-compatible providers and SDKs reuse.

[Data processing agreement (DPA)](https://broadrouter.com/glossary/dpa/)

A data processing agreement (DPA) is the contract between a controller and a processor that governs how the processor handles personal data. GDPR Article 28(3) requires it and lists its contents: documented instructions, confidentiality, security, subprocessor rules, deletion or return of data, and audits.

[Open-weight model](https://broadrouter.com/glossary/open-weight-model/)

An open-weight model is a machine learning model whose trained weights are published for anyone to download and run, usually under a licence with conditions. It differs from fully open-source AI, which also releases the training code and detailed information about the training data, so others can study and reproduce the model.

## Questions

### Does OpenAI train on API data?

Not by default. OpenAI's data controls page says data sent to the OpenAI API "is not used to train or improve OpenAI models" unless you explicitly opt in.

### How long does Anthropic keep API data?

Anthropic's privacy center (updated July 1, 2026) says it automatically deletes API inputs and outputs within 30 days, with exceptions for features like the Files API, zero data retention agreements, Usage Policy enforcement and legal requirements.

### Does Google use Gemini API prompts for training?

On paid services, Google says it does not use prompts or responses to improve its products and logs them for a limited period to detect abuse. On unpaid services, it may use content to improve products, and human reviewers may read it.

### Can I get zero data retention from OpenAI, Anthropic or Mistral?

Yes, with conditions. OpenAI requires prior approval, Anthropic enables it per organisation through sales, and Mistral reviews requests for paid plans. Each covers only stateless endpoints.

### Does Mistral train on API data?

Mistral's help center says it may use free-mode input and output to train its models, with an opt-out. For pay-as-you-go it says customers can opt out at any time; it does not state the default. The opt-out is the Anonymous improvement data toggle under Privacy in the Admin panel.

## Keep reading

-   [Zero data retention LLM API: a 10-question checklist](https://broadrouter.com/blog/zero-data-retention-llm-api-checklist/)
-   [GDPR and LLM APIs: 15 questions to ask your provider](https://broadrouter.com/blog/gdpr-llm-api-checklist/)
-   [Sending sensitive data to an LLM API: redact, minimise, log less](https://broadrouter.com/blog/llm-api-sensitive-data/)
-   [How to choose an LLM API provider: a 20-point checklist](https://broadrouter.com/blog/choose-llm-api-provider-checklist/)
-   [Egypt's data protection law (Law 151 of 2020) and LLM APIs](https://broadrouter.com/blog/egypt-data-protection-law-llm-api/)
-   [How to test an LLM for sensitive-data safety: BroadNet's method](https://broadrouter.com/blog/llm-sensitive-data-safety-evaluation-method/)
-   [Guide: Lock API keys to your servers with IP allow-lists](https://broadrouter.com/guides/api-keys-ip-allow-lists/)

Try it on your real workload. New accounts start with $5 of credit for Gemma and transcription. DeepSeek, GLM and Qwen3.8 need a top-up.

[Get started](https://inference.broadnet.ai/account/?utm_source=broadrouter&utm_medium=referral&utm_campaign=blog-llm-api-data-retention-compared&utm_content=page-cta) ·

**Sources**

-   [OpenAI, Data controls in the OpenAI platform (fetched 2026-11-14)](https://developers.openai.com/api/docs/guides/your-data)
-   [Anthropic Privacy Center, how long do you store commercial API data (updated July 1, 2026; fetched 2026-11-14)](https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-commercial-api-data)
-   [Anthropic, API and data retention ZDR and model-specific retention (fetched 2026-11-15)](https://platform.claude.com/docs/en/manage-claude/api-and-data-retention)
-   [Anthropic Commercial Terms of Service (effective June 17, 2025; fetched 2026-11-15)](https://www.anthropic.com/legal/commercial-terms)
-   [Gemini API Additional Terms of Service (effective March 23, 2026; fetched 2026-11-14)](https://aistudio.google.com/docs/terms)
-   [Gemini API, logs policy (fetched 2026-11-14)](https://aistudio.google.com/docs/logs-policy)
-   [Mistral AI Privacy Policy (effective September 3, 2026; fetched 2026-11-14)](https://legal.mistral.ai/terms/privacy-policy)
-   [Mistral docs, zero data retention (fetched 2026-11-14)](https://docs.mistral.ai/admin/monitor-comply/zero-data-retention)
-   [Mistral Help Center, training on user data (fetched 2026-11-14)](https://help.mistral.ai/en/articles/347617-do-you-use-my-user-data-to-train-your-artificial-intelligence-models)
-   [Mistral Help Center, opting out of training for Studio and API (fetched 2026-11-14)](https://help.mistral.ai/en/articles/455207-can-i-opt-out-of-my-input-or-output-data-being-used-for-training)
-   [BroadNet Inference privacy statement (effective September 2026; fetched 2026-11-14)](https://www.broadnet.ai/inference-privacy)
