DeepSeek: The Price Anchor Starts Moving

The 2026 numbers, with their labels

• 1.6T V4-Pro parameters — MoE flagship, ~49B active per token, 1M context; V4-Flash at 284B/~13B. MIT-licensed open weights. GA August 13, 2026 (V4-Pro-0813).
• +1,100% maximum reported API price increase — Effective August 16, 2026 16:00 UTC, per Caixin and other reporting; the steepest rises fall on previously near-free cache tiers.
• 2× peak over off-peak token pricing — Peak windows 01:00–04:00 and 06:00–10:00 UTC; V4-Pro cache-miss input $1.32 peak / $0.66 off-peak, from $0.435 before.
• 99.80% chat uptime, May–August 2026 — First-party status page; API services 99.73–99.88% in the same window. Contrast: the R1-era launch left most requests hitting server-busy.

From the R1 moment to V4: eighteen months of position change

DeepSeek's R1 moment — January 2025, when a Hangzhou lab backed by the High-Flyer quant fund topped global app stores and knocked roughly six hundred billion dollars off Nvidia's market value in a day, per widespread reporting — set the reference point the whole sector still prices against. Since then the line has stayed on one track: open weights under MIT, from the V3/R1 generation to the V4 preview in April 2026 and V4-Pro's general availability on August 13.

V4-Pro totals 1.6 trillion parameters with about 49 billion active; V4-Flash runs 284 billion with about 13 billion active; both carry a million-token context and both are downloadable. Among the open-weight-first labs this file series covers, DeepSeek has been the most consistent: no revenue-share license, no held-back checkpoints — which is exactly what makes the pricing turn below stand out.

The price pivot: up to 1,100%, and the clock enters the price sheet

With V4-Pro's GA came the announcement: from August 16, API prices rise — up to 1,100% on the tiers that were nearly free — and the sector's first peak/off-peak schedule takes effect, with peak hours at twice the off-peak rate during Beijing business windows. V4-Pro cache-miss input goes from $0.435 to $0.66 off-peak and $1.32 at peak.

Peak pricing is the interesting part: it makes the compute constraint explicit in the price sheet. Demand is being shaped by the clock because the GPUs cannot be. And the direction converges with Zhipu's two 2026 increases — the open-weight labs that undercut the frontier are, one by one, repricing toward sustainability. DeepSeek is the protagonist of this site's weekly price-war report, and this change enters that review cycle as a candidate the moment the new sheet goes live.

Serving under constraint: from server-busy to demand shaping

The R1 launch is also the origin story of the capacity narrative this series keeps meeting: at peak, most requests reportedly hit the server-busy wall. Eighteen months later the first-party status page shows chat at 99.80% uptime and API services between 99.73% and 99.88% over May–August 2026 — stabilized, but visibly not slack.

Put the three data points side by side: Kimi paused new paid signups two days after K3; Manus's transition raised early-signup logic; DeepSeek now prices by the hour. Different labs, one constraint. Our read: in this cycle, serving capacity — not model quality — is the scarce good, and pricing is becoming the rationing mechanism.

On the shockwave timeline: what the anchor moving means

Every wrapper, router and AI feature priced against near-free DeepSeek tokens inherited its economics from the R1 moment. If the anchor moves — and it is moving, by up to 1,100% on some tiers — those downstream cost structures move with it. That is a direct input to the app-category pressure this timeline exists to track.

The open question cuts both ways: either cheap frontier tokens return as capacity catches up, and the price war resumes; or the sector learns that the floor was subsidized all along, and closed-subscription economics get a reprieve nobody priced in. The weekly price report is where this site will keep score.

What this special does not claim

This file separates sourced reporting from open questions. As of publication:

• Parameter counts and uptime figures are company-reported; the 1,100% maximum increase is media-reported pending the live price sheet.
• The Nvidia market-value figure for the R1 moment is widely reported but not audited by this site.
• We have not benchmarked V4 models for this file; it is a sourced company file, not a review.
• Peak/off-peak details may change after the August 16 cutover; the weekly price report tracks the live sheet.

Related reading on this site

• The Model Price War — DeepSeek is this report's protagonist; the August 16 reprice enters its weekly review.
• Zhipu company file — The other open-weight lab raising prices — the repricing is a pattern, not an incident.
• Kimi company file — The capacity constraint from another angle: a two-day pause on new paid signups.
• Claude Academy: Who Bears the Cost of AI Fluency? — Anthropic has launched a free school for learning to work with AI. Its stated framework reaches beyond prompts into delegation, judgment and disclosure. That is a meaningful public resource—and it raises a harder question: when the maker of the disruption also issues the credentials for adapting to it, where does institutional responsibility end and individual responsibility begin?
• Meituan All-in AI: The Execution Costs — Going all-in on AI is easy to announce and hard to govern. The execution bill arrives where strategic urgency meets source provenance, merchant consent and incentives: the less time a team leaves for verification and reversal, the more expensive its speed becomes. This special separates the verified public record from two weak, single-source signals and treats the pattern as a governance problem—not proof that AI investment itself has failed.
• AI Token Monetization: Token Is the New Dollar — At Stripe Sessions, President of Technology and Business Will Gaybrick changed a demo app from a $2 flat fee to $3 per million tokens, then streamed stablecoin payments as each token was consumed. That sequence is more than a billing demo. It shows software moving from seats and monthly access toward metered intelligence: every unit of model work can carry a price, a margin, a fraud risk and a settlement event. Our thesis is that the token is becoming the dollar of AI software—a unit of account for machine work, not legal tender and not a replacement for the US dollar.
• Stripe × OpenRouter: Where Token Monetization Begins — Put the reported $8 billion price aside. Stripe, the payments leader that became a checkout layer for the internet, chose not to buy a model lab but the switchboard between AI applications and hundreds of models. That is the story. Stripe is betting that the most important layer of the AI economy will not only produce intelligence; it will turn intelligence into exchangeable value. Today an LLM token is a billing unit. Tomorrow a model-agnostic AI token could be held, transferred and settled like the generation of digital assets opened by BTC and ETH — representing not digital scarcity or blockchain gas, but a claim on usable intelligence. Axios reports an agreement above $8 billion, while neither company had publicly confirmed it at this update.
• Grok Bot: xAI Gives Every Agent Its Own Computer — Launched in early beta on August 11, 2026, Grok Bot turns the agent from a chat window into a teammate: each Bot gets a persistent cloud computer, signs into the tools you already use and keeps working while you are away. We have been using it since day one. The product is days old and public information is still thin, so this special leads with tested impressions and keeps mechanism facts second — separating what is verified, what is company-stated and what is our read.
• Qwen: The Open-Weight Leader Starts Charging for the Crown — Alibaba's Qwen is the most-downloaded open-weight model family in the world — by company count, more than three billion downloads and over half the open-source market. In August 2026 it shipped its biggest flagship yet, priced it far under US frontier rates, and put its open weights under a revenue-share license for the first time. This file collects the sourced record and our read on what the pivot means.
• MiniMax: The Multimodal Tiger That Doubled on Debut — MiniMax reached the Hong Kong exchange one day after Zhipu and doubled on its first day. But the reason it closes this series is not the listing — it is the product surface. Where the other five files cover text and agents, MiniMax ships video, speech and music at commodity prices, and that points the AI shockwave at a different cohort: creators. This file collects the sourced record and our read.
• A Tribute to Manus — The independent special that anchors this site's digital-labor storyline.

Try DeepSeek

DeepSeek runs a free consumer chat with open-weight models on Hugging Face and API access via its platform.

Open deepseek.com

Official link — no affiliate relationship. If DeepSeek opens an affiliate program, this site will disclose it.

Sources and evidence boundaries

Launch and pricing facts anchor to first-party announcements and major financial press; uptime is the first-party status page; company figures are labeled company-reported.

1. DeepSeek API docs — news — First-party announcement channel, including the V4-Pro-0813 GA notice.
2. Caixin Global — DeepSeek launches V4-Pro, raises API prices up to 1,100% — The GA-plus-reprice report anchoring the price-pivot facts.
3. Quartz — DeepSeek raising API prices from Aug. 16 — Peak/off-peak windows and per-tier rates.
4. DeepSeek status page — First-party uptime record for chat and API services.

Cite this

anti-ai.app, “DeepSeek: The Price Anchor Starts Moving”, https://www.anti-ai.app/specials/deepseek/ (2026-08-15 · 2026-08-17).

Republication policy