What changed
DeepSeek's peak and off-peak V4 API price schedule became active at 16:00 UTC on 16 August. V4-Pro-0813 now lists $0.022/$0.044 per million cache-hit input tokens, $0.66/$1.32 for cache-miss input, and $1.98/$3.96 for output at off-peak/peak rates; V4-Flash uses lower paired rates.
Why it matters
The change moves a scheduled price table into current production economics. A workload's bill now depends materially on model, UTC window, cache state, input volume, output volume, and concurrency rather than a single token price.
What to do
Reprice one representative workflow with actual cache-hit, cache-miss, input, and output tokens across DeepSeek's 01:00–04:00 and 06:00–10:00 UTC peak windows and its off-peak hours, then reconcile the estimate with the next invoice.
Caveat
The first-party page confirms the current rate table, but prices remain changeable and no independent invoice was available during verification. Caching, retries, output length, and UTC execution time can dominate realized cost.