Update: DeepSeek V4.1 Flash goes live today; V4 Pro traffic reroutes
Beta expires today. Confirmed pricing: $0.60/M output off-peak, doubled in new weekday peak windows — and early tests show V4 Pro still wins on dashboards.
Copy markdown
The reroute goes live today — no opt-in
Yesterday's V4.1 Flash beta was opt-in. Today it flips: DeepSeek says that once V4.1 Flash launches (12 PM Beijing / 8 AM Dubai) it will “route all requests for V4 Pro entirely to V4.1 Flash, billing them at the V4.1 Flash unit price” until a V4.1 Pro ships. If you run V4 Pro in production, your model swaps under you today with no code change on your end.
Confirmed price — plus surge hours that double it
Official off-peak rates land at $0.15/M uncached input, $0.60/M output, and $0.003/M cached input. The new wrinkle the beta didn't have: peak hours (Mon–Fri 01:00–04:00 and 06:00–10:00 UTC) double every rate. Run batch jobs off-peak to keep the roughly 3x savings over old V4 Pro output pricing.
Reportedly beats V4 Pro — except on dashboards
DeepSeek claims V4.1 Flash “comprehensively surpassed V4 Pro in performance, cost, speed.” Independent OpenDesign Arena runs back the headline (81.2 vs 72.9 overall) but flag one regression: V4 Pro still leads dashboard-generation tasks, 83.0 to 76.1. No official technical report or benchmark suite has shipped, so treat the clean sweep as reportedly.
Do this before 8 AM Dubai
Snapshot your current V4 Pro prompts, settings and outputs now for a reproducible baseline, then A/B V4.1 Flash on your real workflows — comparing total cost including review time, not just token price. If a chart- or dashboard-heavy task regresses, you'll want that baseline captured before the reroute lands.