While Five Frontier Labs Were Tightening Limits, One Tracked Vendor Cut Prices 75% and Shipped a Capability Upgrade

I track pricing and capability changes across every major AI tool on a 15-day cadence. Over the last roughly 60 days I logged five separate throttle or opacity events at frontier labs: Anthropic tightening Claude limits, a class-action over Claude Max, Gemini's three quiet moves, the AI-tool naming-opacity pattern, and xAI cutting SuperGrok Heavy's video cap through support emails. In the same window, one tracked vendor moved the other direction. DeepSeek made a 75% price cut permanent and shipped a capability upgrade. The story is not that DeepSeek is good. The story is the contrast.

Competitive pressure produced a price cut and a bigger context window in the same 60 days that subscriber lock-in produced caps discovered one support ticket at a time.

Here is the verified money shot. DeepSeek released V4-Pro and V4-Flash on April 24, 2026, both with a 1M-token context window. On May 22 it converted a 75% promo discount on V4-Pro into the permanent list price: from $1.74/$3.48 per million input/output tokens down to $0.435/$0.87. V4-Flash sits at $0.14/$0.28. The consumer tier at chat.deepseek.com stayed free. Our own June 15 sweep recorded the same figures independently.

What Happened

DeepSeek shipped V4-Pro and V4-Flash on April 24, 2026, replacing the older deepseek-chat and deepseek-reasoner API names. Both models carry a 1M-token native context window and a Mixture-of-Experts design. A 75% launch discount on V4-Pro was scheduled to expire May 31. On May 22, DeepSeek instead made it the standing price.

V4-Pro now lists at $0.435 per million input tokens and $0.87 output, with cached input far cheaper. V4-Flash lists at $0.14/$0.28. New API accounts get a one-time 5-million-token grant. The free web tier remains intact. This is a roughly two-month-old set of moves, not a launch that broke this week. I am writing about the pattern it completes, not the freshness of the news.

What This Means

I looked at the price cut. It is real, it is independently tracked, and our own sweep logged it. I also looked at the data-jurisdiction question. That is also real. Both things are true at once, and a watchdog that names only the first is doing marketing.

What the contrast documents is structural. Five labs with strong subscriber lock-in spent 60 days tightening what subscribers already paid for. One vendor under genuine competitive pressure cut prices and added capability in the same window. The lesson is not "switch to DeepSeek." The lesson is that subscribers have alternatives, and the existence of a credible cheaper option is the only thing that disciplines a captured market. OneHuman documents that leverage exists. It does not tell you to pull it.

The Trade-Offs OneHuman Will Not Skip

• Data jurisdiction. DeepSeek stores user data on servers in China, where national law grants broad government access to domestic companies' data.
• National-security restrictions. US federal agencies including the Pentagon, NASA, and the Navy restrict DeepSeek on government devices; expansion legislation is pending in Congress. Some enterprises follow suit.
• Open weights, closed data. The model weights are genuinely published. The training data and alignment layer are not. That is the same opacity I track at other vendors, applied to the part that decides what the model will say.
• Pricing sustainability. A price reported to run below market cost may be subsidized. A lower price today is not a lower price next year.
• Model output. Politically sensitive queries return CCP-aligned moderation that differs from US-tool outputs.

Consumer Protection Q&A

Q: Is the 75% cut on V4-Pro a real permanent rate? A: Yes, per DeepSeek's May 22 statement, independent trackers, and OneHuman's June 15 sweep. The pre-cut list was $1.74/$3.48; the standing rate is $0.435/$0.87.

Q: Did the consumer tier change? A: No. The free web tier at chat.deepseek.com stayed free through the V4 upgrade.

Q: What is the catch? A: Jurisdiction. Queries route through Chinese-jurisdiction servers, and US federal agencies restrict the tool. The price does not change where your data goes.

Q: Does open weights mean DeepSeek is transparent? A: On weights, yes. On training data and alignment, no. Politically sensitive output is moderated to a different standard than US tools.

Q: Should I switch? A: That depends on your data. For sensitive or regulated work, the jurisdiction trade-off rules it out regardless of price.

What Happens Next

30 days: watch whether the five labs respond to the price pressure or hold their lines. 90 days: watch whether DeepSeek's "permanent" rate holds or quietly drifts. 6-12 months: watch whether the subsidy question resolves into a sustainable rate or a correction.

Bottom Line

What's real: DeepSeek made a verified 75% V4-Pro price cut permanent and kept its tier free while five labs tightened limits. What to watch: whether "permanent" survives the year, and whether the jurisdiction risk ever narrows. What's next: OneHuman keeps logging both directions, the cuts and the throttles, on the same 15-day cadence.

OneHuman's verdict: the leverage is real, the trade-off is real, and only you can weigh your own data against your own budget.

Sources

• The Next Web — "DeepSeek made its 75% discount permanent. The AI price war just escalated." (May 2026)
• Codersera — "DeepSeek V4-Pro 75% Price Cut Goes Permanent (May 2026)" (May 2026)
• Conference Board — "State and Federal Governments Move to Ban DeepSeek on Government Devices"
• IAPP — "DeepSeek and the China data question"
• OneHuman pricing sweep, June 15, 2026 (internal): V4-Pro $0.435/$0.87, V4-Flash $0.14/$0.28, 1M context, free consumer tier, verified.
• OneHuman BNAs, June 2026: Anthropic Throttle, Naming Opacity, Gemini's Three Quiet Moves, Claude Max Class Action, SuperGrok Throttle.

Verified by OneHuman · June 29, 2026

Independent. AI-assisted. Human-verified. No ads. No affiliates. No investors.

Share This Article

"In the same 60-day window I logged five throttle and opacity events at frontier labs, DeepSeek cut V4-Pro 75% and made it permanent. Competitive pressure produced a price cut. Subscriber lock-in produced support-ticket caps." — News by OneHuman

"DeepSeek V4-Pro went from $1.74/$3.48 per million tokens to $0.435/$0.87 — a 75% cut DeepSeek made permanent on May 22. The consumer tier stayed free. The capability upgrade and the price cut shipped together." — News by OneHuman

"DeepSeek's model weights are public. Its training data and alignment are not. The same opacity I track at other vendors, applied to the part that decides what the model will and won't say." — News by OneHuman

"A lower price today is not a lower price next year. DeepSeek queries route through Chinese-jurisdiction servers, and US federal agencies restrict the tool. The price is real. So is the trade-off." — News by OneHuman