Thermal imaging works on the principle of acquiring Infrared (IR) radiation emitted by the objects, making them independent of the light source which enables several applications in night-vision, industry, medical, wildlife, etc. However, they demand wide field-of-view to gather sufficient energy at the sensor’s input which hinders the spatial resolution of the acquired images. Thus, hardware limitations of the thermal imaging impose to improve their spatial resolution through software-driven techniques referred as Super-Resolution (SR). This work addresses the computationally efficient image SR for thermal data based on transformer referred as CompTSRTrans. It utilizes transformer model with residual design, a modified transformer encoder to extract high-frequency details from thermal images. The performance of the proposed model is validated across multiple upscale factors (2 and 4) using datasets such as PBVS, FLIR, and KAIST. Thus, the cross-dataset validation further substantiates the model’s robustness, with comprehensive comparisons of PSNR and SSIM across multiple datasets. The quantitative and visual qualitative comparisons with the other existing SOTA networks reveal that the proposed model delivers superior quantitative and visual performance over the other existing SR methods with considerably less number training parameters.