I get ~55 Tok/s on my framework desktop with the 35B A3B q8 model, and so far am also very happy with the coding performance.
did you upgrade to MTP?