DeepSeek V4 Price Increase 2026: New Peak/Off-Peak API Pricing Explained
Quick verdict
On August 17, 2026 DeepSeek moved V4 API pricing to peak/off-peak rates and raised prices. Off-peak Flash is now $0.22/M in and $0.66/M out; using off-peak hours and cache hits keeps costs low.
DeepSeek's ultra-low API pricing has been one of the biggest stories in AI for 2026. On August 17, 2026, that changed: DeepSeek raised V4 Flash and V4 Pro API prices and introduced a peak/off-peak pricing model.
The new rates are no longer the lowest in the industry, but DeepSeek remains far cheaper than GPT-4o or Claude-class models — especially if you schedule workloads around off-peak hours and maximize cache hits.
What Changed
DeepSeek replaced its flat per-million-token pricing with two rate tiers:
- Off-peak hours
- Peak hours (2x off-peak)
Peak hours are 01:00-04:00 and 06:00-10:00 UTC. All other hours are off-peak.
New Pricing at a Glance
DeepSeek V4 Flash
| Billing item | Off-peak | Peak |
|---|---|---|
| Input (cache hit) | $0.007 / 1M | $0.014 / 1M |
| Input (cache miss) | $0.22 / 1M | $0.44 / 1M |
| Output | $0.66 / 1M | $1.32 / 1M |
DeepSeek V4 Pro
| Billing item | Off-peak | Peak |
|---|---|---|
| Input (cache hit) | $0.022 / 1M | $0.044 / 1M |
| Input (cache miss) | $0.66 / 1M | $1.32 / 1M |
| Output | $1.98 / 1M | $3.96 / 1M |
Prices are from DeepSeek's official API pricing page as of August 17, 2026. DeepSeek reserves the right to adjust pricing, so always check the official page.
Old vs New
| Model | Old (flat) | New off-peak (cache miss / output) |
|---|---|---|
| V4 Flash | $0.14 / $0.28 | $0.22 / $0.66 |
| V4 Pro | $0.435 / $0.87 | $0.66 / $1.98 |
Input prices rose roughly 50-60%, and output prices more than doubled for both models. Peak rates are another 2x on top of that.
What It Means for Developers
1. Use off-peak hours
If you can batch heavy workloads into off-peak hours, you pay half the peak rate. This is the single biggest cost lever now.
2. Maximize cache hits
Cache-hit input is now $0.007 per 1M (V4 Flash, off-peak) — about 30x cheaper than cache-miss input. Structuring prompts and context to hit the cache is essential.
3. DeepSeek is still the budget option
Compared to GPT-4o (~$2.50/$10) or Claude Opus-class models, even peak DeepSeek pricing remains far cheaper. The "46x cheaper" claims from earlier 2026 need updating — at off-peak rates, V4 Pro is still roughly 20x cheaper than Claude Fable 5-style pricing.
4. Re-evaluate cost estimates from older articles
Any article or calculator built on the old $0.14/$0.28 or $0.435/$0.87 pricing is now outdated. Budgets should be re-run with the new peak/off-peak rates.
Related Articles
Related Articles
OpenAI Cuts GPT-5.6 Sol API Price by 20%+: What Developers Should Know
OpenAI lowered GPT-5.6 Sol developer API pricing by more than 20% starting August 21, 2026. See the approximate new price, why it matters, and how it changes the AI pricing landscape.
DeepSeek V4 Flash: Official Release With Major Agent Performance Boost (2026)
DeepSeek V4 Flash official release review. 284B MoE model, 1M context, 384K output, and dramatically improved agent benchmarks. Full review with pricing, benchmarks, and use cases.
DeepSeek: Complete Guide to the R1 & V4 AI Models (2026)
An in-depth review of DeepSeek—covering R1 reasoning, V4 Flash and V4 Pro (official release 0813), pricing, and real-world performance. Is DeepSeek still the best value AI in 2026?
DeepSeek Harness: A Hands-On Guide to the Everything-is-a-Plugin Agent Framework
DeepSeek Harness (dsh) v0.1 is an MIT-licensed agent framework where everything is a plugin. Learn how to install it, configure models, choose workspaces, and run agent tasks.