Rio3.5, a locally-trained model, outperforms Qwen on benchmarks

Rio3.5, a locally-trained model, outperforms Qwen on benchmarks

Hacker News·1w·lucasfcosta

Lucas Costa's Rio3.5 model, trained on Rio de Janeiro city government data, exceeded Qwen3.7's performance on recent benchmarks. The result suggests specialized, domain-focused models can compete with general-purpose LLMs—a potentially relevant insight for indie builders working with niche datasets.

Share𝕏Reddit

Related stories