I'm working with a BJT (doesn't really matter what model), and trying to understand the value of gm.

According to first-order theory, gm = Ic/Vt.  gm is really defined as the partial derivative of Ic with respect to Vb.

Now, if I do a sweep of the base voltage, I can plot Ic/Vt and derivative(Ic) on the same graph, but they don't agree for high base voltage (off by a factor of 2).  Ic/Vt does match the "gm" reported by the device fairly well.  I've tried turning off self-heating, and I've checked that there isn't much base current (Vbe - Ib * Rb hardly differs from Vbe) and there also isn't much avalanche current.

What else needs to be considered?

Geoffrey,

  The gm is in series with the Re of the transistor at high
currents so the effective transcondcutance is limited by the
Re of the transistor. In the earlier days of polysilicon
emitters, this effect was one of the concerns about
migrating to polysilicon emitters along with the stability
of the interface.

                                                     Best Regards,

                                                        Sheldon