BACKGROUND: Linezolid plays a crucial role in the first-line treatment of drug-resistant tuberculosis globally. Its prolonged use can lead to neurological and haematological toxicity, highlighting the need for safer oxazolidinones. Delpazolid, a novel oxazolidinone, might be safer. We aimed to evaluate the safety and efficacy of delpazolid and identify an optimal dose. METHODS: PanACEA-DECODE-01 was a prospective, randomised, open-label, phase 2b, multicentre, dose-finding trial done in five tuberculosis trial sites in Tanzania and South Africa. Adults aged 18-65 years, who weighed 40-90 kg, and had newly diagnosed, smear positive pulmonary tuberculosis were randomly assigned (1:1:1:1:1) through centralised allocation, using a probabilistic minimisation algorithm to receive no delpazolid (D0), delpazolid 400 mg once daily (D400), delpazolid 800 mg once daily (D800), delpazolid 1200 mg once daily (D1200), or delpazolid 800 mg twice daily (D800BD), all administered orally for 16 weeks with follow-up to week 52. All participants received bedaquiline (400 mg orally once daily for the first 14 days, then 200 mg orally thrice weekly), delamanid (100 mg orally twice daily), and moxifloxacin (400 mg orally once daily). Randomisation was stratified based on bacterial load in sputum as measured by GeneXpert cycle threshold (<16 vs ≥16), site, and HIV status. The primary efficacy objective was to establish an exposure-response model with the primary endpoint, measured in the modified intention-to-treat population, of change in mycobacterial load measured by time to positivity using the liquid culture mycobacterial growth indicator tube system. A secondary outcome was the time on treatment to sustained conversion to negative sputum culture in liquid media. The primary safety outcome was the occurrence of oxazolidinone class toxicities defined as peripheral or optical neuropathy, incident leukopenia, anaemia or thrombocytopenia, or adverse events in line with tyramine pressor response, all of grade 2 or higher, possibly, probably or definitely related to delpazolid. This study was registered with ClinicalTrials.gov, NCT04550832. FINDINGS: Between Oct 28, 2021, and Aug 31, 2022, 156 individuals were screened for eligibility, 76 of whom were enrolled and randomly assigned to D0 (n=15), D400 (n=15), D800 (n=15), D1200 (n=16), or D800BD (n=15). 60 (79%) of 76 participants were male and 16 (21%) were female. Population pharmacokinetic-pharmacodynamic modelling suggests maximal microbiological activity at a daily total exposure of delpazolid (area under the concentration curve from 0 h to 24 h [AUC0-24]) of 50 mg/L per h; close to the median exposure observed after a 1200 mg dose. This maximal effect was estimated at a 38% (95% CI 4-83; p=0·025) faster decline in bacterial load compared with no delpazolid. In the secondary time-to-event analysis, there was no significant difference in time to culture conversion between treatment arms or exposure tertile. When all delpazolid-containing groups were combined, the hazard ratio for the time to sustained culture conversion to negative, comparing all delpazolid-containing groups with the group without delpazolid was 1·53 (95% CI 0·84-2·76). Two drug-related serious adverse events (one gastritis and one anaemia) occurred in the D800BD group, with high individual AUC0-24. Apart from the anaemia and one event of brief, moderate neutropenia observed at only one visit in the D800 group not in line with the characteristics of oxazolidinone class toxicity, no oxazolidinone class toxicities occurred. INTERPRETATION: The pharmacokinetic-pharmacodynamic modelling results suggest that delpazolid adds efficacy on top of bedaquiline, delamanid, and moxifloxacin; and that a dose of 1200 mg once daily would result in exposures with maximum efficacy. That dose was shown to be safe, raising hope that linezolid toxicities could be averted in long-term treatment. Delpazolid is a promising drug for future tuberculosis treatment regimens and could be widely usable if safety and efficacy are confirmed in larger trials. FUNDING: LigaChem Biosciences, EDCTP2 programme supported by the EU; German Ministry for Education and Research; German Center for Infection Research; Swiss State Secretariat for Education, Research and Innovation; and Nederlandse Organisatie voor Wetenschappelijk Onderzoek.