Terry KS SINGAPORE, 29 SEPTEMBER 2026 – Applied artificial intelligence research lab AutoTrust AI Pte. Ltd. has set a rapid pace in the global race toward recursive self-improvement, posting leading results across four public benchmarks designed to test self-improving artificial intelligence models throughout September 2026.
Operating as a seed-stage laboratory that has raised under USD 10 million to date, the Singapore-headquartered startup successfully competed against heavily funded global competitors, including entities that have raised seed rounds as large as USD 300 million and USD 650 million. The company revealed that its agentic research platform, ScienceGuru, secured first place on the Autoresearch@Home benchmark, achieved the top validated entry on MedARC NanoPath v2 upon posting, and logged the fastest reported completion times on two separate GPT-2 training speedruns.
Powered by AutoTrust’s proprietary Guru Turbo foundation models, ScienceGuru autonomously proposed, implemented, and tested each technical recipe, with human researchers solely setting overarching objectives and allocating compute resources. The system executed a standardized research loop across all tests, surveying literature, fusing top community concepts, optimizing memory layouts, and running standard hardware evaluations before publishing verified code publicly.
According to AutoTrust co-founder and chief executive Daniel Tang, recursive self-improvement represents the defining competitive frontier for next-generation artificial intelligence development. Tang highlighted that a lean team operating from Southeast Asia successfully claimed top benchmark metrics while utilizing a fraction of the capital deployed by major international laboratories.
Beyond serving as benchmark accomplishments, ScienceGuru forms the core layer of AutoTrust’s enterprise software stack, which functions as an agentic operating system for model building and research execution. The company is currently making the software available across macOS and Windows operating systems while leveraging the framework to assist commercial clients in training customized sovereign artificial intelligence models using internal enterprise data.