But of course toml is superior to both because it has comments and no significant white space and the comma at the end of lists is optional.
JSON doesn't have "integers" and "floats", it has "numbers" which are are expressed in decimal scientific notation. There's no guarantee an implementation will let you distinguish what was parsed, and the numerical precision and range of interoperable JSON is undefined.
The bottom line is JSON is a risky proposition for financial or scientific data exchange.
And before you say JSON numbers are de-facto IEEE754 doubles, bare in mind that most JSON implementations can't roundtrip them properly.[0] A symptom of everybody wanting to write a toy implementation.
[0] https://rawgit.com/miloyip/nativejson-benchmark/master/sampl...
TBH, common JSON implementations have their problems too. Integer size and integer/float (de-)serialization tend to be somewhat interesting at times. I've even seen implementations serializing JavaScript objects as JSON (e.g., {foo: 2} instead of {"foo": 2})