CoralOS is an open infrastructure project that's basically "Kubernetes for AI agents". They're building infra for the stuff between your agents and production - registry, runtimes, security, orchestration. The benchmark: Last year they tested their multi-agent system on the GAIA benchmark (General AI Assistants) and got a 34% higher score compared to comparable setups using similar-sized models. GAIA is one of the most challenging benchmarks for AI agents. It evaluates whether systems can complete real-world, multi-step research and problem-solving tasks that require reasoning, tool use, and information gathering. Humans score around 92% on it, while most models struggle to get anywhere near that. What makes this interesting: Instead of using one massive model, they got better results by orchestrating multiple smaller models together. Their approach is what they call "horizontal scaling" - getting specialized agents to collaborate.…