Loading the catalogue…
Loading the catalogue…
Nexusflow was a small Palo Alto startup that released open-weight fine-tunes (Athene, NexusRaven, Starling) of Llama and Qwen base models, but was acquired by NVIDIA around mid-2025 and its founders now work there — its own website is offline (SSL failure) and there is no active team maintaining, patching, or making compliance commitments for the models. EU regulated customers should treat any Nexusflow-branded model as an orphaned, US-jurisdiction open-weights artifact with no GDPR/AI Act infrastructure, rather than a vendor-supported product; self-hosting the weights inside the EEA mitigates data-transfer risk but not licensing or support gaps.
Nexusflow was acquired by NVIDIA (approx. mid-2025) and no longer operates as an independent entity; its founder is now confirmed working at NVIDIA, and the company's website has been unreachable (SSL hostname mismatch) across three separate research checks in 2026. There is no active team maintaining, patching, or providing compliance support for Nexusflow-branded models.
US-incorporated entity, now effectively folded into NVIDIA (also US), confirming full CLOUD Act exposure with no EU legal presence identified anywhere in company records.
No EU AI Act compliance statement, GPAI Code of Practice participation, or training-data transparency summary could be located for Nexusflow; targeted searches surfaced no Nexusflow-specific material, only generic third-party guidance.
Nexusflow's last-known privacy policy addressed only US state privacy laws (California CCPA, Nevada) with no GDPR-specific provisions, DPO contact, or EU data subject rights language; the page is now unreachable.
Model weights are distributed under mixed custom licences (Nexusflow Research License, Nexusflow community license) that restrict commercial/hosted redistribution more than a standard permissive open-source licence, and the issuing entity is now defunct, creating legal uncertainty for enterprise deployment.
Stav’s assessment
Editorial assessment, not legal advice. Stav's risk ratings, scores, and verdicts are our own analysis of publicly available information and may be incomplete or out of date. Verify independently before making compliance or procurement decisions.
Nexusflow published detailed model cards, benchmark methodology, and training lineage (e.g. Athene-V2 built from Qwen2.5-72B, Athene-70B from Llama-3-70B) on HuggingFace, along with accompanying blog posts.
NexusRaven-V2's model card explicitly documents that it was instruction-tuned without using proprietary LLM-generated data, giving enterprise customers clearer provenance information than many peers.
Nexusflow maintained a public GitHub org, Discord community, and HuggingFace leaderboard for its NexusRaven function-calling benchmark during its active period (2023–2024).
Published safeguards & certifications