Major breakthroughs by frontier labs over the last 3 months came from scaling test-time compute with agent swarms. I've been building an open-source equivalent for the past year and would love feedback from the community. [P]

For benchmarks I couldn't test this on frontier models or even large open models because it will be way too expensive. I directly only tested this on Gemma4-35B & Qwen-3.8-27B. They spent around 200M tokens / problem and