All stories
Free LLM API Hub Ecosystem Community reports

Free API does not mean private: this hub adds prompt-training policy fields

Version 2.11.0 records what providers say about training on prompts and retaining data, with a source and check date. An empty field means unknown, not a privacy verdict.

By AITrending2 min read
Sources (4)

Why now The English v2.11.0 release was published October 13 at 13:31:32 UTC and surfaced in the October 15 discovery pass. Its new policy fields are a dated release event within 48 hours, not a claim that the directory or listed providers launched today.

Put the data bargain next to the free tier

Free LLM API Hub has added a practical question to its directory of no-cost AI offers: what does the provider say happens to your prompts? Its October 13 v2.11.0 release introduces a data_policy object with prompt-training, retention, source and check-date fields.

The distinction matters when a prototype moves from harmless sample text to material a team would not want reused. A free allowance describes the bill; it does not, by itself, establish a provider’s data-handling policy. The new fields make those separate questions easier to inspect together.

The release explicitly leaves a policy empty when the project cannot cite a provider statement. That does not mean the provider trains on prompts, and a recorded no-training statement is not a blanket guarantee that nothing is logged or retained. Readers still need the linked policy, its scope and its date.

A dataset for humans and tools

The MIT-licensed community project offers a browser explorer, provider comparisons and public JSON exports. Developers can consume the dataset without a model key; actually using a listed API still depends on that provider’s credentials, signup conditions, quotas and terms. It is a directory and data source, not an inference gateway or a new model.

The release also documents OpenAI-client and LiteLLM configurations, a weekly change feed and a score-driven top list. It bundles changes since v2.9.1 because the intervening tag lacked a release-notes page. An MCP package is mentioned, but marked unpublished; the notes do not establish an available hosted MCP service.

For reproducible tooling, the README recommends pinning a tagged snapshot rather than silently following the moving dataset on main.

Dated evidence, not a permanent stamp

The tagged methodology defines verification as checking core offer details against provider documentation on a recorded date. It expressly excludes a real-time guarantee or endorsement. The ranking also contains editorial inputs, and its stability component is unmeasured without a sufficient probe history.

AITrending checked the release, methodology and policy records. This report covers the directory’s new documentation workflow, not an independent audit of every provider’s privacy practices or a recommendation to send sensitive data to a particular free endpoint.

Related reading