It seems that pytest.approx repr is optimized for very high or very low tolerances, but is not as easily readable as it could be for values in between.
Some examples:

• pytest.approx(42, abs=1) repr is 42 ± 1.0e+00, which could be just 42 ± 1.
• pytest.approx(5, rel=0.01) repr is 5 ± 5.0e-02, but could be 5 ± 0.05 or maybe even 5 ± 1%
• pytest.approx(24000, abs=500) repr is 24000 ± 5.0e+02 and could be 24000 ± 500
• removing accuracy example: pytest.approx(1500, abs=555) repr is 1500 ± 5.6e+02 and could be 1500 ± 555

In these number ranges using the exact number instead of the exponential notation is both more accurate and easier readable for humans. Please adapt pytest.approx to switch it's repr style according to the input value ranges.