Rapid tests for HIV testing are essential tools to achieve the 90-90-90 target of the World Health Organization. Many tests are available, some directly from websites. Evaluation of the performance of rapid tests, under close to real-life usage, is therefore needed to ensure accurate diagnosis in the context of the recommendation for their more widespread use. Method: Nine third- (3G) or fourth-generation (4G) rapid screening tests or self-tests (two bought on websites), were evaluated on an extensive panel of 200 HIV-negative and 312 HIV-positive samples, representative of a wide variety of clinical situations and HIV genetic diversity. A whole blood reconstitution protocol was designed to simulate real-life usage of these tests in community-based and private settings. Findings: The specificity was high (98.5–100%) and sensitivity excellent (100%) for samples from patients chronically infected with the pandemic strains. The performance for infrequent situations with a major epidemiological and clinical impact, such as infection with divergent viruses or primary infection, was highly variable, depending on the test. One of the two 4G tests allowed detection of additional positive samples from early stages of infection, whereas the second (sold as a 4G test on a website) corresponded in reality to a 3G test. Interpretation: Our study showed that not all tests are equal for the detection of major HIV variants or early stages of HIV infection; adding the detection of specific p24Ag improved the latter point. This study also showed, for the first time, that buying through web-based vendors can be risky, due to the varying performance of the tests and questionable sales practices. Our results are of particular importance in the context of the increasing use of rapid tests in an “outside laboratory” settings.