DeepSeek V4.1 Flash and V4 Pro: API Routing versus Weights

An API name and a downloadable checkpoint are not always the same model. DeepSeek's V4.1 Flash announcement says requests to deepseek-v4-pro route to V4.1 Flash from September 14, 2026 at 04:00 UTC, until V4.1 Pro launches. The separate V4 Pro model card describes the released checkpoint.

Select the route you need

DeepSeek describes V4.1 Flash as a 552B MoE model with native visual understanding. Its announcement reports API changes and peak/off-peak pricing. Check the current API table and route before budgeting rather than applying an old V4 Pro rate to the new alias.

The V4 Pro card describes a 1.6T total / 49B active model with a million-token context and marks its licence MIT. Those checkpoint details do not establish the model currently served by a third-party provider.

Before migrating

Record the provider, exact model identifier, checkpoint if applicable, date, supported inputs and relevant settings. Run a small representative task in a test environment and inspect both the result and the usage record. For a comparison, keep the task and acceptance criteria fixed.

If an older request still works, that only establishes compatibility at that moment; it does not prove the underlying model remained unchanged.

Sources checked October 30, 2026. Varwo has not downloaded the weights, tested a serving stack, made paid calls or reproduced the provider's performance claims. General access context is in our DeepSeek listing.