OPEN LINK ↗
// 2h agoMODEL RELEASE
IplanRIO releases Rio-3.5-Open-397B MoE model
Developed by IplanRIO, Rio-3.5-Open-397B is an open-weight Mixture-of-Experts model featuring 397 billion parameters post-trained on Qwen 3.5 with a 1,010,000-token context window. The release utilizes a training-free inference framework called SwiReasoning, though it has sparked debate over whether it is a merge of existing models.
// ANALYSIS
While a city government releasing a massive 397-billion-parameter open model with a million-token context window is a fascinating milestone for public-sector AI, the community's skepticism regarding it being a weighted merge suggests we should await independent, rigorous verification before crowning it a true Qwen 3.7-killer.
- –Public sector sovereign AI: It shows an intriguing shift of municipal/governmental bodies building and sharing sovereign AI models, leveraging existing massive foundational models like Qwen.
- –Merge vs. Fine-tune transparency: Community suspicion that the model is a weighted merge (e.g., using Nex-N2 Pro and Qwen 3.5) highlights the persistent challenge of verifying originality and provenance in open-source AI.
- –SwiReasoning efficiency: If the claim of a training-free inference framework dynamically managing reasoning steps is validated, it could offer a highly token-efficient paradigm for other LLMs.
// TAGS
rio-3.5-open-397biplanrioqwenmoellmopen-sourceswireasoning
DISCOVERED
2h ago
2026-06-14
PUBLISHED
4h ago
2026-06-14
RELEVANCE
7/ 10
AUTHOR
lucasfcosta