Sign Language Translation (SLT) can be used to facilitate communication between Deaf and hearing people. Some SLT tasks involve glosses, such as gloss-to-text and text-to-gloss. In this research, we explore the performance of gloss-to-text and text-to-gloss translation models in translating BISINDO gloss to Indonesian text and Indonesian text to BISINDO gloss. We fine-tune four pre-trained models, namely IndoNanoTS-base, MVP, MVP-multi-task, and NLLB-200-distilled-600M, on our new dataset, namely TVRI BISINDO 7 DAYS, to perform the translation. The TVRI BISINDO 7 DAYS dataset consists of 1644 BISINDO and Indonesian sentence pairs. We use SacreBLEU, chrF++, METEOR, and ROUGE-L scores to evaluate models' performance. The results of this research show that NLLB-200-distilled-600M achieves the highest SacreBLEU, chrF++, METEOR, and ROUGE-L scores for gloss-to-text translation tasks. NLLB-200-distilled-600M achieves the highest SacreBLEU and chrF++ scores, while IndoNanoTS-base achieves the highest METEOR and ROUGE-L scores, for text-to-gloss translation tasks. © 2025 IEEE.