What happened
IplanRIO released "Rio-3.5-Open-397B" as an original 397B model. Nex-AGI subsequently alleged the model was a 0.6 Nex / 0.4 Qwen3.5-397B-A17B element-wise merge, citing consistent weight tensors across all 60 layers and the model identifying as "Nex" 79% of the time. IplanRIO updated its model card, acknowledging the model was built via a merge of Nex-N2-Pro and Qwen3.5-397B-A17B, attributing the initial discrepancy to an incorrect upload of a base merged version instead of a final distilled model.
Why it matters
Transparency in open-source model development is critical for maintaining trust and verifying intellectual property. Founders and investors evaluating AI projects must scrutinise training claims, as IplanRIO's initial misrepresentation and subsequent correction highlight the need for verifiable model provenance. Procurement teams require clear documentation of training methodologies to ensure compliance and performance expectations. This incident underscores the importance of independent verification, particularly as new open-source models like DeepSeek V4 continue to challenge frontier AI capabilities.
