DeepSeek Didn’t Cut Prices. It Raised Them and Called it a Discount.
Last Updated on August 24, 2026 by Editorial Team
Author(s): allglenn
Originally published on Towards AI.
DeepSeek’s ‘discount’ will cost you 4x more. Here’s how to fight back.
Before August 16, 2026, DeepSeek-V4-Pro output cost a flat $0.87 per million tokens. After August 16, the cheapest version of that same output, off-peak, no exceptions, costs $1.98 per million. That’s a 128% increase, marketed as a discount because DeepSeek is comparing it to the peak rate instead of to what you paid last week. Land in one of the two daily peak windows and output runs $3.96 per million: more than four times the old price.
The article explains that DeepSeek’s “discount” comes from a peak/off-peak billing shift plus automatic prompt caching, not from genuinely cheaper calls: all rates went up, and “up to” figures usually refer to cache-hit input during peak windows. It lays out the actual peak windows (UTC) and the pricing table framing the change, then shows why peak time doesn’t magically preserve savings when cache and time-of-day multipliers apply independently. To claw back most of the difference without abandoning DeepSeek’s ecosystem, it proposes three engineering levers: route requests based on true complexity (send routine/high-volume work to Flash and reserve Pro for the hardest tasks), keep the system prompt’s prefix identical so caching reliably hits, and move non-urgent jobs out of peak hours using UTC-aware scheduling. The piece includes a worked example to rebuild the new cost basis, guidance for testing and instrumentation (cache-hit token logging, boundary checks, dry-run scheduling logs, spend alerts), and highlights common mistakes that destroy caching or cause teams to budget against “50% off” marketing. Finally, it summarizes who benefits most, what trade-offs and operational complexity come with the optimization, and gives a practical checklist for updating production routing, prompting, and budgeting based on recent billing logs.
Read the full blog for free on Medium.
Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.
Published via Towards AI
Towards AI Academy
We Build Enterprise-Grade AI. We'll Teach You to Master It Too.
15 engineers. 100,000+ students. Towards AI Academy teaches what actually survives production.
Start free — no commitment:
→ 6-Day Agentic AI Engineering Email Guide — one practical lesson per day
→ Agents Architecture Cheatsheet — 3 years of architecture decisions in 6 pages
Our courses:
→ AI Engineering Certification — 90+ lessons from project selection to deployed product. The most comprehensive practical LLM course out there.
→ Agent Engineering Course — Hands on with production agent architectures, memory, routing, and eval frameworks — built from real enterprise engagements.
→ AI for Work — Understand, evaluate, and apply AI for complex work tasks.
Note: Article content contains the views of the contributing authors and not Towards AI.
Related posts
Recent Posts
