Model optimizer is more focusing on the model version rather than the framework version, so shouldn't the question be:
For specific model, doesn't its version validated by model optimizer can be trained by Tensorflow 0.12?
Looks like I was wrong, you can check the following file, in Rel 3, this should be at line 51, it seems support TensorFlow >=1.2.0:
<OpenVINO install dir>/deployment_tools/model_optimizer/mo/utils/versions_checker.py
Let me know if it helps.