We determine the radii and masses of 293 nearby, bright M dwarfs of the CARMENES survey. This is the first time that such a large and homogeneous high-resolution (R>80 000) spectroscopic survey has been used to derive these fundamental stellar parameters. We derived the radii using Stefan-Boltzmann's law. We obtained the required effective temperatures $T_{\rm eff}$ from a spectral analysis and we obtained the required luminosities L from integrated broadband photometry together with the Gaia DR2 parallaxes. The mass was then determined using a mass-radius relation that we derived from eclipsing binaries known in the literature. We compared this method with three other methods: (1) We calculated the mass from the radius and the surface gravity log g, which was obtained from the same spectral analysis as $T_{\rm eff}$. (2) We used a widely used infrared mass-magnitude relation. (3) We used a Bayesian approach to infer stellar parameters from the comparison of the absolute magnitudes and colors of our targets with evolutionary models. Between spectral types M0V and M7V our radii cover the range $0.1\,R_{\normalsize\odot}<R<0.6\,R_{\normalsize\odot}$ with an error of 2-3% and our masses cover $0.09\,{\mathcal M}_{\normalsize\odot}<{\mathcal M}<0.6\,{\mathcal M}_{\normalsize\odot}$ with an error of 3-5%. We find good agreement between the masses determined with these different methods for most of our targets. Only the masses of very young objects show discrepancies. This can be well explained with the assumptions that we used for our methods.

We report parallax measurements for 70 ultracool dwarfs (UCDs). Using both literature values and our sample, we report new polynomial relations between spectral type and M$_{JHK}$. Including resolved L/T transition binaries in the relations, we find no reason to differentiate between a "bright" (unresolved binary) and "faint" (single source) sample across the L/T boundary. Isolating early T dwarfs, we find that the brightening of T0-T4 sources is prominent in M$_{J}$ where there is a [1.2 - 1.4] magnitude difference. A similar yet dampened brightening of [0.3 - 0.5] magnitude happens at M$_{H}$ and a plateau or dimming of [-0.2 - -0.3] magnitude is seen in M$_{K}$. Comparing with evolutionary models that vary gravity, metallicity, and cloud thickness we find that a near constant temperature of 1200 $\pm$100 K along a narrow spectral subtype of T0-T4 is required to account for the brightening and color magnitude diagram of the L-dwarf/T-dwarf transition. Furthermore, there is a significant population of both L and T dwarfs which are red or potentially "ultra-cloudy" compared to the models, many of which are known to be young indicating a correlation between enhanced photospheric dust and youth. For the low surface-gravity or young companion L dwarfs we find that 8 out of 10 are at least [0.2-1.0] magnitude underluminous in M$_{JH}$ and/or M$_{K}$ compared to equivalent spectral type objects. We speculate that this is a consequence of increased dust opacity and conclude that low-surface gravity L dwarfs require a completely new spectral-type/absolute magnitude polynomial for analysis.