DeepSeek V4 Pro GA: API Prices Reported Up to 1,100%

Verdict: DeepSeek has executed the hike it only planned a week ago. V4 Pro went GA (per DeepSeek's changelog), and per Reuters and Caixin, API prices now rise as much as 1,100% with peak/off-peak billing replacing the flat rate. Both DeepSeek V4 Pro and V4 Flash stay Recommended today — but the pricing math behind that rating just changed, and both are flagged for re-review.

On Aug 13, DeepSeek released the official version of its flagship DeepSeek-V4-Pro and sharply raised API prices, with some rates up 1,100%, while introducing peak/off-peak pricing — per Caixin Global, which cites a company statement also reported by Reuters. This is the execution of the "significant price increase" DeepSeek signaled on Aug 5–6, which we covered in last week's fyi. Then it was a plan; now it is a number.

What happened

• V4 Pro is GA. The flagship left its ~4-month preview as the official V4-Pro build. Caixin reports it scores 53 on the Artificial Analysis Intelligence Index — tying GLM-5.2.
• Prices rose sharply. Reuters reports increases spanning 50% to 1,100% depending on model, token type, and time of day.
• Flat pricing is dead. Peak/off-peak billing now applies to both V4-Pro and V4-Flash. TechNode reports peak hours are 9 a.m.–noon and 2–6 p.m. Beijing time, effective mid-August.

The rates in our directory — $0.435/$0.87 for V4 Pro and $0.14/$0.28 for V4 Flash — are the pre-hike flat numbers. New peak/off-peak rates take over in mid-August (outlets report Aug 16–17). Check DeepSeek's live pricing page before quoting budgets.

Why it matters

DeepSeek was the pricing floor. Its "7–107× cheaper than Western frontier" positioning didn't just win volume — it set the reference point every other provider is measured against. When the floor rises, headroom opens for Claude Sonnet 5 ($3/$15), GPT-5.6 Sol ($5/$30), and Grok 4.5 ($2/$6) — exactly as OpenAI has been cutting prices (Luna −80%, Terra −20% on Jul 30).

The two Recommended verdicts were written on the old math. V4 Flash's "cheaper than any Western frontier-adjacent API by 7–107×" and V4 Pro's "3× cheaper than Fable 5 per completed task" are now stale premises. The models themselves are unchanged — it is the economics that moved. A verdict re-review is warranted, but that is the verifier's call, not something this draft sets.

The open weights still hedge the blow. Both models are MIT-licensed. V4 Flash self-hosts on a single H100; V4 Pro on 8× H100s. A 1,100% API increase tilts the ROI math hard toward self-hosting for high-volume teams — the weights remain free.

What changes for you

1. Rebuild your cost model. The $0.14/$0.28 and $0.435/$0.87 flat rates are gone — pull DeepSeek's current peak/off-peak card before budgeting.
2. Re-run the self-host ROI. If your DeepSeek API bill is material, the hardware case just got dramatically better.
3. Watch the directory. Both DeepSeek verdicts are flagged for re-review on this price move.

FAQ

When do the new prices take effect?
Aug 16, 16:00 UTC (Aug 17 Beijing time), per DeepSeek's changelog. Peak hours are 9 a.m.–noon and 2–6 p.m. Beijing time; off-peak covers everything else.

Does self-hosting avoid the increase?
Yes. Both models are MIT-licensed and self-hostable — V4 Flash on a single H100, V4 Pro on 8× H100s. The API hike strengthens the self-hosting ROI case; the weights themselves stay free.

Are the DeepSeek verdicts changing?
Not today — both stay Recommended. But the "7–107× cheaper" and "3× cheaper than Fable 5" premises were written on the old flat rates, so both models are flagged for a re-review of the pricing math.