From the Changelog:

- If a model has integer variables, but the model is already integer optimal in the simplex fase,
  then it was reported as suboptimal feasible.
- get_objective, get_variables, get_ptr_variables, get_constraints, get_ptr_constraints, get_primal_solution
  reported 'not a valid basis' when presolve is active and the model is entirely solved by presolve.
- presolve on a model with SOS variables sometimes went wrong.
- presolve on a model with SOS variables where infeasibility is detected crashed.
- read_bas could fail when not all constraints had names or had names like default variable names.
- A crash could occur with set_row(ex) in rowmode.
- The lp format has been extended with a free section to define free variables.
- bfp_GLPK and xli_MathProg are now compiled against GLPK 4.13
- fixed bug in the pseudocost logic that can blow up subsequent pseudocost values in that
  branch and make them almost random.
- In some rare cases a memory overrun could occur when constraints are added after a previous solve.
- Made the copy_lp routine work. Note that information about the optimisation of the original model
  is not copied (at this time). Only the model is.
- Fixed a bug in the hashing routines that had only influence in some rare cases in the
  MATLAB, O-Matrix, Scilab, Octave, Python drivers.
- coldual sometimes worked on a uninitialised variable value with unpredictable results.