ChinaModelAPI

News / Model Watch · Pricing Update

Pricing update Weekend = all-day off-peak Effective Aug 23
2026-08-23 · Model Watch · announced Aug 22 evening, live Aug 23 00:00 Beijing

DeepSeek Drops Weekend Peak Pricing: Saturdays & Sundays Now All-Day Off-Peak

Less than a week after peak/off-peak billing took effect (Aug 17), DeepSeek simplified the weekend: from August 23, 00:00 Beijing time, Saturdays and Sundays no longer have peak windows — every weekend hour bills at the off-peak rate. Today is the first all-day off-peak Sunday.

Direct answer

Weekends are now flat-rate off-peak on the DeepSeek API — announced the evening of Aug 22 per IT之家/新浪科技 and 科创板日报, effective Aug 23 00:00. That means deepseek-v4-pro output at ¥13.5/M (vs ¥27 weekday peak) and v4-flash at ¥4.5/M — all day Saturday and Sunday. Weekday windows are unchanged: peak 9:00–12:00 & 14:00–18:00 Beijing, everything else (including nights) off-peak.

The full weekend rate card (effective Aug 23)

Per 1M tokensWeekend / off-peakWeekday peakNote
v4-flash · input cache-hit¥0.05 ($0.007)¥0.10 ($0.014)rowspan applies to vision-exp too
v4-flash · input cache-miss¥1.5 ($0.22)¥3.0 ($0.44)
v4-flash · output¥4.5 ($0.66)¥9.0 ($1.32)
v4-pro · input cache-hit¥0.15 ($0.022)¥0.30 ($0.044)
v4-pro · input cache-miss¥4.5 ($0.66)¥9.0 ($1.32)
v4-pro · output¥13.5 ($1.98)¥27.0 ($3.96)weekend saving: ¥13.5/M

Rates cross-checked against the official pricing page (USD) and the CN page (¥); the weekend rule itself comes from the Aug 22 announcement per IT之家/新浪/科创板日报 — the EN page footnote had not yet been updated at publication.

What it means for scheduling

  • Weekends become the batch window. All-day half price means long agent runs, eval sweeps, and cache warming cost exactly 50% of weekday-peak — a 100M-output-token v4-pro batch costs ¥1,350 vs ¥2,700 at weekday peak.
  • Weekdays: keep the two peak windows in mind. Beijing 9–12 & 14–18 remain 2× pricing; nights are already off-peak.
  • The direction of travel is load-shaping. First peak/off-peak split (Aug 17), then weekend simplification (Aug 23) — expect the pricing page to keep evolving; watch the official changelog.

Primary sources

FAQ (2026)

What changed?

From Aug 23, weekends have no peak windows — all-day off-peak. Weekday windows unchanged; pre-change fees settle under old rules.

Weekend rates?

v4-pro ¥13.5/M output (half of ¥27 peak); v4-flash ¥4.5/M. Same as weekday off-peak — now guaranteed all Saturday and Sunday.

vision-exp too?

Platform-wide billing rule; vision-exp is priced identically to v4-flash, so yes — weekend all-day off-peak applies the same way.

When announced / live?

Announced Aug 22 evening (IT之家/科创板日报, following user feedback); live Aug 23 00:00 Beijing — six days after the original mechanism.

Scheduling advice?

Batch jobs and cache warming → weekends (all-day half price) or weekday nights. Latency-sensitive daytime traffic avoids weekday 9–12 & 14–18 Beijing. A 100M-token v4-pro output batch saves ¥1,350 by moving from weekday peak to a weekend.

Related guides