Graph-based multi-agents smash long-context benchmarks–89% MMLU-Pro on 8B models

(github.com) by FeelTheAGI2 | Feb 16, 2026 | 1 comments on HN