Our CTO and I tried getting it to write some relatively challenging Swift as a benchmark example and it just repeatedly botched it. It would produce something close to working code, but kept insisting on using libraries that didn't have support for what it was trying to do with them, which was also an issue with 3.5.