The objectives of this study were to compare and model feed intake responses to temperature across species and to assess opportunities to use cross-species (CS) data to parameterize models when species-specific (SS) data were limited. Literature searches were conducted to identify studies reporting intake during climate-stress treatments compared with 1 or more thermoneutral treatments. The resulting data set comprised 614 treatment means for 108 studies on livestock responses to heat or cold stress. An ANOVA was conducted on the CS data set to identify the effects of species, temperature and species by temperature interactions on intake as a fraction of thermoneutral intake (FFI). Four models were derived from the CS data set and SS root mean squared prediction error (RMSPE) and concordance correlation coefficients (CCC) of these models were compared with models of the same form derived from SS data sets. Models used explanatory variables for (1) duration of exposure; (2) mean temperature; (3) low and high temperatures; and (4) difference between low and high temperatures and duration of exposure. An additional model accounting for temperature and stage of production was derived from the SS data. ANOVA demonstrated that the species by temperature interaction did not have a significant effect on FFI (P = 0.162). Across species, intake decreased with temperature. Notably, all species demonstrated a constant decrease in intake across the thermoneutral zone indicating the previous assumption of constant intake during thermoneutrality may be incorrect. The CS-derived models had marginally lower RMSPE and higher CCC when compared with those derived from the SS data sets. The model fit with production data had the lowest RMSPE and highest CCC within the study. When compared over areas with notable knowledge gaps, using CS models often had reduced RMSPE and improved CCC when compared with SS models. Although fitting models based on SS data allows for incorporating unique covariates, like level of production, fitting responses based on CS data can help to improve model estimates when knowledge gaps exist.