Rio de Janeiro "homegrown" LLM alleged to be a model merge

Category: Pulse
Status: discussion
Sources: GitHub issue, HN discussion
Summary: A GitHub issue and a 321-point Hacker News thread allege that a model promoted as a Rio de Janeiro city-government LLM is a merge or fine-tune of existing open-weight models rather than a model trained from scratch, with a separate thread questioning benchmark claims that it beats Qwen. The provenance allegation rests on community inspection of the published weights and configuration and is not confirmed by the project. Treat both the "homegrown" framing and the rebuttal as unverified.
Comments: Commenters pointed to tokenizer and config artifacts they read as evidence of an upstream base model, and warned that self-reported benchmark wins from undisclosed merges are easy to game; none of this is independently confirmed.