Why Did DeepSeek Raise API Prices by Up to 1,100%
Right After China's Open-Weight AI Blitz?

01

Timeline: what happened, and when

The easy mistake is treating "1,100%" as the whole invoice. Start with the dates, then keep billing dimension, license type, and benchmark source separate.

Zoom out: on Jul 30, OpenAI cut prices on its cheapest tier (GPT-5.6 Luna, down 80%), then on Aug 6–7 made Luna the free default with unlimited text chats. While Chinese labs were raising prices and opening flagship weights, US labs were cutting prices and going free at the consumer layer — at the same time. That is two sides of the same pricing fight.

Six misreads to drop before you quote a headline

1. 01

   Treat 1,100% as the whole bill. That figure is peak-hour cache-hit input, the tier that started nearest to free.

2. 02

   Ignore Beijing peak windows. Peak hours are 9am–12pm and 2pm–6pm Beijing time. The announcement is asking you to shift load.

3. 03

   Call any open checkpoint Apache 2.0. Qwen3.8-Max ships under a custom license. Muse Glimmer is a different bet.

4. 04

   Repeat the geo-ban rumor. The published license has no US / EU / UK / Korea territorial clause.

5. 05

   Read GLM-5.3 as a new foundation model. Same 743B base as 5.2. The jump is post-training RL scale.

6. 06

   Assume Chinese model still means cheapest model. Off-peak DeepSeek still undercuts Claude Opus 5. Luna and Qwen international pricing now undercut DeepSeek.

03

Breaking down the three strategies

Three labs. Three plays. One question underneath: when usage scales faster than GPUs and pretraining returns flatten, who still holds pricing power.

DeepSeek: time-of-day pricing is a capacity problem

The easiest misread is "China's cheapest model finally caved to margin pressure." The structure reads more like the opposite: a company making compute constraints visible in the price sheet. Flat, always-cheap pricing worked as a customer-acquisition tool as long as GPU capacity kept pace. Once usage grew exponentially and capacity did not, something had to become explicit. "Encouraging more flexible workload scheduling" is corporate-speak for "peak-hour compute is now scarce, please shift your load yourself."

One detail international coverage mostly missed: at peak hours, DeepSeek's official API price is now higher than several third-party resellers (GMI Cloud, Novita, and others currently list V4 Pro below DeepSeek's new peak rate). The assumption that the official API is always the cheapest way to run DeepSeek has been broken for the first time.

Alibaba: open weights buy mindshare; a custom license protects the ceiling

Alibaba did two things at once. It published the full 2.4T-parameter checkpoint for free download, and it attached a custom license — not the permissive Apache 2.0 used for smaller Qwen models. Any Model-as-a-Service or AI Work Assistant business earning over $50 million in any 12-month period must negotiate a separate commercial license. Products with 100M+ monthly active users or $20M+ in monthly revenue must prominently display the model name.

Give away the weights to win developer mindshare, especially internationally. Keep pricing leverage over the handful of companies that can build a competing inference business on top. That is a materially different bet than Meta's Muse Glimmer, which ships under unrestricted Apache 2.0.

GLM-5.3: no new base model, just a bigger post-training bet

The most interesting fact is the method: same 743B-parameter base as GLM-5.2, no retraining, and a roughly 6x jump on Terminal-Bench 3.0 (4.6% → 28.3%) from scaling reinforcement-learning environments in post-training. As pretraining scaling laws show diminishing returns, post-training RL scale is becoming an independent performance lever with a much lower cost floor than retraining a new foundation model. Mid-tier labs without OpenAI-scale compute budgets can still close the gap on agentic and coding benchmarks.

05

What is disputed, why it matters, and three numbers you can cite

What's disputed or unverified

• The 1,100% headline is technically accurate but misleading without context. It applies only to peak-hour cache-hit input. Output pricing rose 350%. Different outlets have quoted different tiers as if they were the whole story.
• Claims that Qwen3.8-Max runs on Alibaba's in-house Zhenwu M890 chips (and "Pangu AL128" supernodes), reported by several Chinese financial outlets, have not been independently confirmed by Alibaba's own technical documentation or third-party benchmarks. Treat this as vendor-adjacent, unverified reporting.
• GLM-5.3's reported discovery of a "serious vulnerability" in Cursor comes from VentureBeat and Zhipu's own disclosure. Specific technical details have not been made public. Read it as a vendor-sourced, not independently audited, security finding.
• Reports that China's Ministry of Commerce may be preparing retaliatory export controls on AI / semiconductor technology are speculative and sourced to unconfirmed media reports, not an official announcement.

Two price wars running in parallel

Over roughly the past month, China's top labs have shipped major releases at a pace domestic financial media has started calling "three model updates a week" (一周三更) — DeepSeek, Alibaba, and Zhipu, plus Moonshot's Kimi K3 (open-weighted Jul 16, 2.8T parameters) and MiniMax H3. Chinese coverage broadly frames this as Chinese open-weight releases forcing a global repricing of the AI industry.

US labs are running the opposite play at the consumer layer: OpenAI cut prices 80% on its cheapest tier (Jul 30) then made that model free and unlimited a week later (Aug 6–7); Google shipped a coding-focused model at half the price of its three-week-old predecessor (Aug 13). Chinese labs open-weight flagships and introduce tiered, higher pricing on the compute-constrained top end. US labs race toward free and cheap at the consumer end.

There is also a geopolitical layer worth naming carefully. Moonshot's Kimi K3 open-weighting in July already drew US security scrutiny. Alibaba choosing this window to open-weight a 2.4T flagship has been read by some analysts as a move to lock in international mindshare and a "technological parity" narrative before any potential regulatory tightening. That is an informed interpretation, not a confirmed fact — and it is easy to miss if you only read English-language product posts.

Citeable figures (as of publication)

• DeepSeek V4-Pro peak output: ¥27.0 per 1M tokens, up 350% from ¥6.0. Cache-hit input: ¥0.025 → ¥0.30, about 1,100%.
• Qwen3.8-2.4T-A95B: 2.4T total / 95B active, MoE 512 experts (10 routed + 1 shared), 262,144-token native open checkpoint.
• GLM-5.3 Terminal-Bench 3.0: official 28.3% (5.2 was 4.6%), still below GPT-5.6 Sol 34.6% and Claude Fable 5 at 33.7%.

Sources: DeepSeek's official pricing announcement, cross-checked against Wall Street CN, IT Home, AIGC.cn, and V2EX; Alibaba's official Qwen repositories (Hugging Face / ModelScope) and South China Morning Post on license terms; Zhipu (Z.ai) GLM-5.3 technical page, plus VentureBeat and StableLearn; Meta AI Research's official blog; Chinese financial outlets (Yicai, Sohu Finance) on release pacing. Verify the latest official pricing and license terms before republishing. Details flagged above as unverified have not been independently confirmed.

Peak/off-peak API pricing makes "always-on flagship inference" more expensive and harder to forecast. Self-hosting a 2.4T MoE is a datacenter problem, not a laptop problem. Resellers can shave the peak rate, but parking production agents, Xcode builds, and long sessions on a shared inference queue still leaves you with weak isolation. For a more stable production environment that fits iOS CI/CD and AI-agent automation, NodeMini's Mac Mini cloud rental is usually the better fit — dedicated Apple Silicon, node-level billing, no fight with Beijing peak hours for the same GPU invoice.