120 agentic runs: the 35B coder post-trains, and the model that knew it was lying27 July 2026·11 minsLocal-Ai Qwen Llama-Cpp Benchmarking Agents Homelab
90 agentic runs, zero failures, and one invented person26 July 2026·15 minsLocal-Ai Qwen Llama-Cpp Benchmarking Agents Homelab