Open Inference has launched, providing researchers and individual users with free inference access to selected open source models through OpenRouter. The goal is to accelerate open research and optimize the inference experience with a "quality-first" approach. The official website and announcement explain that users can directly access the service by enabling the appropriate settings on OpenRouter. This service is intended for free research-based access, not for commercial or programmatic use.
The privacy policy states that the service records submissions and outputs (including minimal technical logs), does not store IP addresses, cookies, or account identifiers, and automatically and manually desensitizes the data before publication. The data is shared with the community in an anonymous, reusable format for research, security assessments, and service improvements. Because user identities are not tracked, submissions cannot be individually withdrawn; users who do not wish to be used should avoid submitting sensitive information.
Frequently Asked Questions
Q: What does Open Inference provide?
A: Free access to a number of open source models through OpenRouter for open research and evaluation, with an emphasis on inference quality.
Q: How is the data processed?
A: Prompts and outputs will be recorded and de-identified, and may be released in a masked form for research and security assessment.
Q: Can it be used commercially or in batches?
A: No. The terms are limited to personal, non-commercial, and non-programmatic use, and are in the research preview stage.
Q: How do I get started?
A: You can call it after checking the instructions on the Open Inference page and enabling trainable/publicly disclosed free endpoints and providers in the privacy settings of OpenRouter.
Q: Can I delete the submitted data?
A: Because personal identities are not tracked, individual submissions cannot be withdrawn; please do not submit sensitive personal information.
