We let models localize into 16 languages. How we made it read native.

The announced update expands the model's language capabilities to include sixteen distinct languages, allowing it to process input and generate output directly in those languages. This shift means the model itself performs the localization work rather than relying on a downstream translation service. The change is presented as a native reading ability, implying tighter integration of language understanding within the model's core. For developers, the primary benefit is a simplified stack: applications can call the model and receive localized results without orchestrating separate translation APIs. This can reduce latency, lower operational costs associated with third-party translation services, and improve consistency between the model's understanding and its output. The native approach also opens the door to more nuanced language handling, as the model can leverage its internal representations for each language. Adopting the new capability will require updating the model version used in production and ensuring that language detection or selection logic routes requests to the appropriate localized path. Teams should allocate time for regression testing across all sixteen language