DeepSeek Doubled Its Rate Card and Still Leads Daily Use

DeepSeek's price-leader era just got more expensive. Official API docs put new DeepSeek V4 Pro rates live at 16:00 UTC on August 16, 2026: cache-miss input moves from $0.435 to $0.66 off-peak and $1.32 at peak, and output jumps from $0.87 to $1.98 off-peak and $3.96 at peak. Cache-hit input goes from $0.003625 to $0.022 off-peak and $0.044 at peak. The model id is unchanged — DeepSeek-V4-Pro-0813 — and arena still lists deepseek-v4-pro at 1458 Elo. The invoice is what changed.

Peak hours are 01:00–04:00 and 06:00–10:00 UTC, seven hours billed at twice the off-peak rate. helloai lists the off-peak $0.66/$1.98 figure the same way it lists Gemini 3.1 Pro's standard $2/$12. That is still the cheapest tracked frontier API. Muse Spark 1.2 remains $1.25/$4.25, Qwen3.8-Max $2/$6, Gemini 3.1 Pro $2/$12, Claude Opus 5 $5/$25, and Claude Fable 5 $10/$50. Off-peak DeepSeek is about half of Muse on input and output. Peak DeepSeek ($1.32/$3.96) almost meets Muse on input and still undercuts it on output.

The August 7 rebalance sold DeepSeek as an order-of-magnitude undercut of Western mid-tier rates. That sentence is now false. Versus Muse it is a 2x story, not a 10x story. Teams that built agent loops around $0.003625 cache hits will feel the 6x to 12x cache-hit hike more than the headline input bump. Peak windows also sit on a slice of US evening and Asian morning, when a lot of interactive coding agents run. The MIT license and 1M context did not change; the set-it-and-forget-the-bill pitch did.

helloai is not dropping DeepSeek. Honest Daily Use still belongs to the cheapest tracked API with a 1M window, and no other row undercuts $0.66/$1.98. Grok 4.6 is still $2/$6 with a 500K window and a preliminary 1464 Elo — too fresh for the two-week bar, and it does not win on price. Gemini 3.7 Flash shipped August 13 at an introductory $0.75/$3.75, but Flash does not replace 3.1 Pro on hard reasoning, and 3.5 Pro is still missing from Google's pricing page.

The next weekly pass will be about whether anyone else moves. If Muse or Qwen hold their rates, DeepSeek keeps the daily-use slot even after a painful hike. If they cut, or if Grok's arena sample matures above the floor with a unique agent story, the August 7 price-shock admit becomes a one-month footnote. Until then, /api/recommend still routes high-volume, non-Mythos work to DeepSeek — just not at the invoice you budgeted on August 15.