DeepSeek V4 Pro traffic silently routes to V4.1 Flash from September 14: upgrade with a price cut

From 12:00 Beijing time on September 14, 2026, DeepSeek's previous flagship DeepSeek V4 Pro API requests are silently routed to the DeepSeek V4.1 Flash model released September 10 and billed at the V4.1 Flash rate, letting users pick up an upgrade and a price cut without code changes. The routing change aligns with V4.1 Flash's release cadence. V4.1 Flash uses a Causal-Encoder-Decoder architecture with 552B total MoE parameters and asymmetric activation of 8B on input and 16B on output, scoring roughly 2.5 to 3 times V4 Pro on Terminal-Bench 3.0 and 4.0 and lifting DeepSWE v1.1 from V4-Flash's 54.4 to 74.2. V4.1 Flash retains the peak-and-off-peak pricing structure: off-peak rates are RMB 0.02 per million cached input tokens, RMB 1 per million uncached input tokens, and RMB 4 per million output tokens, with peak rates (workdays 09:00-12:00 and 14:00-18:00) at exactly double. The model unifies V4 Pro's fast, expert, and vision modes into a single smart mode that auto-detects query complexity, activates vision on image input, and adjusts effort to task difficulty; WorkBuddy, CodeBuddy, and OpenCode have already integrated V4.1 Flash as official launch partners. DeepSeek has simultaneously cut the V4.1 Flash list price and silently moved V4 Pro traffic onto V4.1 Flash, compressing both the upgrade experience and unit-economics gap at the same time. The company is also preparing a STAR Market IPO with CITIC Securities as lead underwriter, with the second-round pre-money valuation reportedly around RMB 500 billion. Industry observers read the move as further intensifying price competition in the domestic LLM API market and putting pressure on peers still pricing at V4 Pro levels, while delivering a simultaneous upgrade and price cut on top of V4 Pro.