I was using analytical Jacobian. I also will not obey your evaluation that a failed lsqcurvefit does not deserve a 1 star. The NIST solutions are tested using different methods (hence the standard deviation is also available), so failing even one of them means there are codes out there that performs better, even using the poor starting condition.
Furthermore, 3 of the 5 failure cases are observed data. They are not some cooked up example designed to cause the optimization to fail. Therefore, if used in real world examples, there is a possibility that it will not work. I am simply letting everybody know a caveat cantor that the program solution should be treated with a grain of salt, even if the sum of squares dropped by 8 orders. Although as the website says, passing all tests is not a pre-requisite to be a good optimization software, but don't you think not passing the tests from an agency that establishes industry standards would not be considered as "industry standard"?
I admit giving it 1 star is too aggressive. I do this to get your attention to respond. Indeed it worked. Unfortunately I can't go back and change it. Because this program works for all easy to intermediate difficulty problems, it deserves a 4 star.