NextFin News — A 27-billion-parameter local large model developed by Shanghai-based StartLux ranked second overall in the China Academy of Information and Communications Technology’s Trusted AI Large Model Benchmark MCP special test, according to a recently published inspection report.
StartLux-V1.0-27B-Preview scored 39.25, placing ahead of DeepSeek-V4-Flash-0731 (284 billion parameters) and Step-3.7-Flash (198 billion parameters). It trailed only the 1.6-trillion-parameter DeepSeek-V4-Pro and outperformed the same-size Qwen-3.6-27B by 5.34 percentage points. The model ranked first or tied for first in several individual tasks, including location navigation, browser automation and financial analysis.
The MCP special test evaluates multi-tool collaboration, complex task execution and real-environment interaction across six task categories plus an overall assessment. StartLux said the model was post-trained from a Qwen3.6-27B base using the company’s own multi-dimensional optimization method and an AI-trains-AI approach. The company plans to release its first-generation local intelligent solution later this year and said the model can already run on consumer-grade personal computers.
Explore more exclusive insights at nextfin.ai.
