Abstract Introduction This study assessed the bone cuts accuracy of a robotic-assisted system for total knee arthroplasty (TKA) that was recently upgraded. Materials and methods Three orthopaedic surgeons planned and executed TKA on 24 sawbones. Bone cut accuracy was assessed using CT scans, comparing the planning and the actual bone cuts in all six degrees-of-freedom. Results The root-mean-square (RMS) values were below 2 mm or 2° for all error types, except for the medio-lateral position (2.4 mm) and internal-external rotation (2.3°) of the left tibias. The maximal amplitude of the 288 errors (6 degrees-of-freedom * 2 bones * 24 knees) was observed in tibial external rotation (3.2°). Most error types reported a bias, with limited variations among knees. Conclusions The errors were in the same range as those of the prior version of the system, suggesting that the improvements brought by the system upgrade were not obtained at the expense of accuracy.