Let me explain what I'm after:
- all examples are taken from a large corpus of contemporary text (the COBUILD series is now using Bank of English)
- the senses for each entry are sorted by frequency, most frequent first
- the senses are based solely on a concordance based on the corpus
- the definitions are complete stentences without tons of abbreviations
- it's stuffed with collocations (words that occur together very often, like spot -> the right spot, quiet spot)
- there are synonyms and antonyms for high frequency words
- the frequency of words are marked directly, not hidden in an appendix
- it uses IPA, not homegrown pronunciation-symbols (probably common in Non-English dictionaries anyway)
- any irregular forms are shown together with the entry, not hidden in an appendix
Lexicographically, the first COBUILD dictionary heralded a paradigm shift, and I was very lucky to find a copy when I was learning English.
I've been looking for something like a COBUILD-style dictionary for Geman and French for a very long time. I've started to suspect that it maybe takes too many resources to build one (a corpus with few enough rights limitations and a trained staff and several years for the assembly), and that's why there aren't any obvious candidates for other languages.