Prime Intellect has released Prime Agent, an open-source coding harness, available now on the company's GitHub repository.
The distinguishing claim: Prime Agent can revise parts of its own operating setup while it works — a self-modifying harness rather than a fixed scaffold around a model. It is aimed at developers working with modern open and closed frontier models, and the company frames it two ways at once: as a coding assistant, and as a runtime for long-horizon autonomous work.
That is close to the extent of what has been published. No benchmark numbers, supported-model list, or license details surfaced alongside the release, and coverage so far is limited to a single release round-up — treat the long-horizon autonomy framing as vendor positioning until the repo and independent runs say otherwise.
Sources: TestingCatalog