Singapore's Agnes AI Releases Free Agnes 2.5 Models, Claiming Frontier-Level Coding Scores

Singapore-based Agnes AI, developed by Sapiens AI, has published internal benchmark results for its new Agnes 2.5 Pro and Agnes 2.5 Flash models, positioning them as free, frontier-competitive systems aimed at developers and agent builders. The company says Agnes 2.5 Pro scored 82.7 on SWE-bench Verified and 78.7 on SWE-bench Multilingual, while Agnes 2.5 Flash improved on the prior 2.0 Flash generation across every benchmark it reports, with its largest gains on the SWE Atlas coding evaluation.