Volcengine Ark’s docs describe a limited-time change to model deduction rates for Agent/Coding Plan personal subscribers when they use listed models. It only changes how hard those named models hit the plan quota. It does not change other models in the package, Harness, how quotas refresh, or the plan’s usage limits. It is not a cut to the monthly subscription price.

The promotion turns on and off at the times on the calendar—no application needed. Volcengine says it may change the rules or dates; go by the latest official docs page.

What’s on the calendar (official windows)

  • kimi-k3: ended at 23:59:59 on October 25, 2026. During the window, Coding Plan available volume for Kimi-K3 was described as comparable to the Agent Plan half-rate (「5折」) deduction period.
  • kimi-k3.1-pro: October 24 through November 9, 2026, 23:59:59. Same comparison: Coding Plan available volume is described as comparable to the Agent Plan half-rate (「5折」) period.
  • deepseek-v4.1-flash and deepseek-v4.1-flash-economy: through December 30, 2026, 23:59:59. Deduction rate at half the usual rate on top of the existing coefficient—officially “5 折”—so usable quota is described as 200% of the original amount for those models.
  • Auto mode: through November 8, 2026, 23:59:59. In Coding Plan, Auto’s deduction coefficient is 1; it can route to kimi-k3, with a much higher kimi-k3 share overnight (00:00–08:00).
  • Also still listed as ongoing: kimi-k2.8-preview through December 14, 2026, 23:59:59 (available volume compared to an Agent Plan 60%-of-list / 「6折」 period). Several older GLM and DeepSeek/Kimi windows are already marked ended on the same page.

The page’s lead line says discounts can go as low as 2.5 折 across the campaign; each model’s rule in the table is different—not every row is a flat 50% cut.

How much quota these windows yield in practice has not been measured on this site—for reference only. Feedback can go through the site-owner contact link in the footer.

Related reading: Kimi Adds K3.1 Pro.