Work versus precision for computing the matter power spectrum P(k) with SymBoltz and CLASS. SymBoltz is approximation-free and run with several implicit ODE solvers. CLASS is run with approximations on/off, its implicit/explicit evolvers ndf15/rk, and precision settings in Appendix B, but tight coupling is always approximated. For each setup, P(k) is computed by integrating perturbation k-modes with ODE solver tolerances 10⁻²-10⁻⁹ that controls the adaptive stepping. For each tolerance, we record the best of 3 runtimes and the L₂-compressed error $(\sum _{i=1}^N(P(k_i)/\overline{P}(k_i)-1{)}^2/N{)}^{1/2}$ relative to high-precision spectra P(k) from both codes with approximations off, tolerance 10⁻¹⁰, and FBDF/ndf15. Both codes solve a w₀w_aCDM model with equal parameters, N = 114 equal wavenumbers k_i, multipole cutoff l_max = 16 for all species, default sampling of massive neutrino momenta, and parallelization over k. Times are from a Linux laptop with an Intel i7-12800H CPU.