Abstract: Unstructured tetrahedral meshes hold significant application value in complex geometric modeling, yet most existing methods require prior surface mesh generation before volume mesh construction, often leading to 3D mesh generation failures or excessive scale in high-precision simulations of large complex models. To overcome limitations in current techniques, such as low efficiency, poor robustness, and implementation complexity, this paper proposes an octree-refinement-based unstructured tetrahedral mesh generation algorithm. The algorithm adopted an improved root tetrahedral configuration ensuring shape consistency among initial grid units while maintaining quality. Surface fitting was achieved through two operational phases: mesh shape adjustment and mesh splitting. Test results demonstrated the algorithm's tolerance to certain geometric defects, exhibiting a significantly higher proportion of high-quality elements (0.9–1.0 metric range) than other methods, while reducing mesh generation time by over 80% for million-element meshes compared to conventional methods. This approach provides an efficient high-quality mesh generation solution that overcomes traditional limitations, offering new perspectives for research and applications in unstructured tetrahedral mesh generation.