each word is only 11 bits (211=2048) so each 3 words represent 33 bits of entropy except the last word that is part entropy part checksum.
When you take the seed phrase as a whole, then yes, each word encodes 11 bits of data. What I meant was that for every additional 3 words extra in length that you use, you gain an additional 32 bits of entropy to your seed phrase, because there is 1 bit of checksum per 32 bits of entropy.
it is closed source and nobody knows what it does in the background with your seed or what bugs it has.
Good point. Until I read that I had forgotten about the critical bug where Coinomi were found to be sending user entered seed phrases to Google's spell checking service unencrypted.