## DeepSeek API Price Hike — Review & Before/After Comparison DeepSeek **is raising API prices for V4-Flash and V4-Pro** and introducing **peak/off-peak (time-of-day) billing**. Per the official docs, new rates take effect at **16:00 UTC on August 16, 2026** (i.e., Aug 17 Beijing time — Reuters pegs it as Aug 17); peak hours are 01:00–04:00 and 06:00–10:00 UTC, with off-peak rates at exactly **half of peak rates** ([DeepSeek pricing docs](https://api-docs.deepseek.com/quick_start/pricing/); [Reuters via Yahoo Finance](https://finance.yahoo.com/technology/ai/articles/deepseek-raises-api-pricing-v4-115712902.html)). ### Current vs. New pricing (per 1M tokens) **deepseek-v4-flash** (current: cache-hit $0.0028 · cache-miss $0.14 · output $0.28) | Item | Current | New off-peak | New peak | Off-peak multiple | Peak multiple | |---|---|---|---|---|---| | Input (cache hit) | $0.0028 | $0.007 | $0.014 | 2.5× (+150%) | 5× (+400%) | | Input (cache miss) | $0.14 | $0.22 | $0.44 | ~1.6× (+57%) | ~3.1× (+214%) | | Output | $0.28 | $0.66 | $1.32 | ~2.4× (+136%) | ~4.7× (+371%) | **deepseek-v4-pro** (current: cache-hit $0.003625 · cache-miss $0.435 · output $0.87) | Item | Current | New off-peak | New peak | Off-peak multiple | Peak multiple | |---|---|---|---|---|---| | Input (cache hit) | $0.003625 | $0.022 | $0.044 | ~6.1× (+507%) | ~12.1× (+1,114%) | | Input (cache miss) | $0.435 | $0.66 | $1.32 | ~1.5× (+52%) | ~3.0× (+203%) | | Output | $0.87 | $1.98 | $3.96 | ~2.3× (+128%) | ~4.6× (+355%) | This matches Reuters' summary that the new rates are **50% to ~1,100% above current prices**, depending on model, token type, and time of use — the worst case being **V4-Pro cache-hit tokens during peak hours (+1,114%)**, and the mildest being V4-Pro cache-miss input off-peak (+52%) ([Reuters via Yahoo Finance](https://finance.yahoo.com/technology/ai/articles/deepseek-raises-api-pricing-v4-115712902.html)). ### Even further back (pre-V4 baseline) For context, before the V4 lineup (as of a **March 15, 2026** archive of the same page), `deepseek-chat` / `deepseek-reasoner` (DeepSeek-V3.2) cost **$0.028 cache-hit / $0.28 cache-miss / $0.42 output** per 1M tokens ([archived pricing page](https://web.archive.org/web/20260315/https://api-docs.deepseek.com/quick_start/pricing/)). So: - V3.2 → current V4: input dropped ~10× for cache-hit on Flash ($0.028→$0.0028) and output dropped ~1.5× ($0.42→$0.28). - New peak V4 prices land well *above* the old V3.2 prices for output (Flash peak $1.32 vs $0.42 = ~3.1×), while off-peak cache-miss input ($0.22) is still slightly below the old $0.28. ### Key takeaways 1. **Biggest relative hit: cache-hit tokens** — the discounted cache tier is the one being re-priced hardest, especially for V4-Pro at peak (12×). 2. **Peak hours are a steep surcharge**: running during 01:00–04:00 / 06:00–10:00 UTC doubles the off-peak rate, so shifting batch workloads off-peak effectively halves the new bill. 3. **V4-Flash cache-miss input stays cheapest on the roster** ($0.22 off-peak, $0.44 peak), but output costs ~2.4–4.7× current. 4. Model names remain the same (`deepseek-v4-flash` / `deepseek-v4-pro`); the underlying versions were updated to V4-Flash-0731 and V4-Pro-0813 ([DeepSeek API docs](https://api-docs.deepseek.com/quick_start/pricing/)). *Sources opened: [DeepSeek Models & Pricing (live)](https://api-docs.deepseek.com/quick_start/pricing/), [Wayback capture of pricing page (2026-03-15)](https://web.archive.org/web/20260315/https://api-docs.deepseek.com/quick_start/pricing/), [Reuters via Yahoo Finance (2026-08-13)](https://finance.yahoo.com/technology/ai/articles/deepseek-raises-api-pricing-v4-115712902.html).*