For independent builders
Find a route that fits the way you work.
Check how an offer is accessed, what the provider says about its limits, and which terms to read before code leaves your machine.
Better choices start with better records.
FreeBench helps independent developers understand free AI coding-agent access and inspect workload evidence for exact model-and-harness setups.
For agent experimenters
Inspect the exact setup behind a result.
Read the model, access route, harness version, workload and evidence. A model-only claim is too broad to choose an agent setup.
Editorial rule
Keep access facts and performance evidence apart.
Provider-reported
What access currently looks like
Offers, stated limits and setup terms link to their source and the date FreeBench checked them.
FreeBench measured
What happened on a task
Only a reviewed harness run with a deterministic grade and replayable evidence can support a task result.
Privacy / visitor measurement
We measure navigation, not your code.
First-party aggregates count public page views, navigation, discovery actions, offer search and filter use, and outbound-link destinations by approved host. Search terms, query values, full external URLs, credentials and project content are not sent. A random per-tab identifier supports rate limiting and daily session counts; its limiter record expires after 48 hours without activity. FreeBench uses no third-party analytics pixels, and tracking does not block navigation.