We make forecasts for the impact a future "midband" space-based gravitational wave experiment, most sensitive to $10^{-2}- 10$ Hz, could have on potential detections of cosmological stochastic gravitational wave backgrounds (SGWBs). Specific proposed midband experiments considered are TianGo, B-DECIGO and AEDGE. We propose a combined power-law integrated sensitivity (CPLS) curve combining GW experiments over different frequency bands, which shows the midband improves sensitivity to SGWBs by up to two orders of magnitude at $10^{-2} - 10$ Hz. We consider GW emission from cosmic strings and phase transitions as benchmark examples of cosmological SGWBs. We explicitly model various astrophysical SGWB sources, most importantly from unresolved black hole mergers. Using Markov Chain Monte Carlo, we demonstrated that midband experiments can, when combined with LIGO A+ and LISA, significantly improve sensitivities to cosmological SGWBs and better separate them from astrophysical SGWBs. In particular, we forecast that a midband experiment improves sensitivity to cosmic string tension $G\mu$ by up to a factor of $10$, driven by improved component separation from astrophysical sources. For phase transitions, a midband experiment can detect signals peaking at $0.1 - 1$ Hz, which for our fiducial model corresponds to early Universe temperatures of $T_*\sim 10^4 - 10^6$ GeV, generally beyond the reach of LIGO and LISA. The midband closes an energy gap and better captures characteristic spectral shape information. It thus substantially improves measurement of the properties of phase transitions at lower energies of $T_* \sim O(10^3)$ GeV, potentially relevant to new physics at the electroweak scale, whereas in this energy range LISA alone will detect an excess but not effectively measure the phase transition parameters. Our modelling code and chains are publicly available.