The scientific research agent ClawsGO Science, developed by BitMiracle AI Lab, leverages Claude Opus 5.5 as its core decision-making engine and integrates 136 GPT-6 Astra models as execution units. Additionally, it utilizes the Lean theorem prover kernel to rigorously verify each step of its reasoning process. In a remarkable achievement, the system resolved a long-standing conjecture in chemical reaction networks—originally proposed in 1974—within just 11 days. The conjecture posits that for reversible reactions occurring in a closed system with equal total quantities, the system will inevitably converge to the same equilibrium state, regardless of the initial reactant composition. The complete proof has been meticulously documented in 83 publicly accessible Lean files.

While AI has previously demonstrated groundbreaking capabilities in mathematics—solving decades-old unsolved problems—natural science research demands specialized software, extensive databases, and robust computational infrastructure. The cloud workstation paired with ClawsGO Science comes pre-loaded with multidisciplinary research tools and databases, ensuring stable execution of long-term scientific tasks while comprehensively recording every stage of the research process.

This fully automated scientific research agent is now officially available. The development team continues to explore advancements in persistent scientific research agents, verifiable AI reasoning frameworks, and next-generation AI-driven scientific infrastructure.