DeepSeek’s September 10 V4.1-Flash announcement combines an architecture claim with a practical model-identity change. For teams comparing costs, the immediate issue is knowing which model a name refers to before interpreting a lower price or a performance result. DeepSeek’s release notice
DeepSeek describes V4.1-Flash as a 552-billion-parameter mixture-of-experts model with native visual understanding. Its causal encoder–decoder design uses 8 billion active parameters for input and 16 billion for output, according to the company. DeepSeek also claims that the KV cache needs one-quarter of the previous generation’s HBM and one-eighth of its SSD storage. These are vendor architecture and efficiency statements, not independently reproduced measurements.
The official name is deepseek-flash. DeepSeek’s current pricing page identifies that name as V4.1-Flash and says the retired V4-Flash names are still accepted but served by the successor at Flash rates. That is a direct reason the release remains relevant on October 2: an existing name can lead to a newer model. Current model and pricing documentation
A separate part of the rollout is less clear. The release notice says all deepseek-v4-pro requests would route to V4.1-Flash from September 14 until V4.1-Pro launches. The current pricing page still lists deepseek-v4-pro as V4-Pro-0813 with its own rates. Those descriptions do not establish one consistent current state. This article therefore does not claim that every V4-Pro request presently uses Flash.
Pricing also depends on timing: both documents describe off-peak rates as half the peak rates. A cost comparison needs the model identity and billing period, alongside the amount of generated and cached work. Architecture scale alone cannot predict the bill.
My recommendation is to treat version and alias documentation as part of a model purchasing decision, especially when a successor replaces earlier names. V4.1-Flash offers an efficiency proposition worth examining, while the inconsistent V4-Pro descriptions require clarification. The release supplies a reason to reassess model choice; it does not supply an independently verified universal performance or savings result.