A poll of experts predicts a split market by 2028 for coding AI, with some expecting frontier models to maintain a high premium and significant token share due to generalization, while others foresee non-frontier models handling most high-volume coding tasks with smaller premiums. Several respondents anticipate a hybrid approach where frontier labs offer both premium and cheaper models, or where a single 'frontier' query internally blends smaller and larger models for efficiency.
The future token volume and pricing power for AI coding models remains contested, implying varied compute demand for different model tiers.