💰钛媒体•Stalecollected in 70m
DeepSeek Implements Permanent Price Cuts
💡DeepSeek's permanent price cut is reshaping the unit economics of AI. See if your infrastructure costs can be optimized.
⚡ 30-Second TL;DR
What Changed
DeepSeek officially shifts to a permanent low-price model
Why It Matters
This aggressive pricing forces other LLM providers to re-evaluate their unit economics and potentially triggers a broader price war in the AI infrastructure sector.
What To Do Next
Re-calculate your projected inference costs using DeepSeek's new pricing to determine if migrating your production workloads is financially viable.
Who should care:Founders & Product Leaders
Key Points
- •DeepSeek officially shifts to a permanent low-price model
- •Analysts link the pricing strategy to a $10 trillion market opportunity
- •Amazon executives are analyzing the cost-efficiency behind DeepSeek's pricing
🧠 Deep Insight
Web-grounded analysis with 14 cited sources.
🔑 Enhanced Key Takeaways
- •DeepSeek's permanent price reduction for its V4 Pro model is a 75% cut, making a previous promotional offer permanent that was originally set to expire on May 31, 2026.
- •The new permanent pricing for DeepSeek V4 Pro is set at approximately $0.003625 per million input tokens and $0.87 per million output tokens, positioning it to significantly undercut competitors.
- •This aggressive pricing strategy is partly enabled by DeepSeek's optimization for Huawei's Ascend 950 AI chips, reducing its reliance on NVIDIA hardware affected by U.S. export restrictions.
- •DeepSeek's move is seen as a direct challenge to Western AI firms like OpenAI and Google, aiming to capture market share by offering a more affordable alternative for enterprise and power users who consume millions of tokens daily.
- •The company's decision to lock in the discount one month after launching the V4 models suggests a strategic prioritization of market share over immediate per-unit revenue, emphasizing a 'cost-effective 1M context length' era.
📊 Competitor Analysis▸ Show
| Provider | Model | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|---|
| DeepSeek | V4 Pro (New Permanent Pricing) | $0.003625 | $0.87 |
| DeepSeek | V3.2 | $0.28 | $0.42 |
| OpenAI | GPT-5 | $2.50 | $10.00 |
| Anthropic | Claude Opus 4.7 | $5.00 | $25.00 |
| Gemini 3.5 Flash | $0.15 | $0.60 |
🛠️ Technical Deep Dive
- DeepSeek-V3 and DeepSeek-R1 models utilize a Mixture-of-Experts (MoE) architecture, comprising 671 billion total parameters with only 37 billion activated per token, which enhances efficiency during inference and training.
- Key architectural innovations include Multi-head Latent Attention (MLA), designed to optimize attention operations and reduce memory consumption by compressing key-value pairs into a low-dimensional latent space.
- DeepSeek-V3.2 further incorporates DeepSeek Sparse Attention (DSA), which achieves a 70% reduction in computational complexity compared to standard attention mechanisms through learned sparsity patterns.
- The models employ FP8 mixed precision for training, which doubles compute efficiency and halves memory usage compared to BF16, contributing to lower training costs.
- The DeepSeekMoE component, specifically in V3, uses 256 routed experts and 1 shared expert, with each token dynamically interacting with 8 specialized experts plus the single shared expert, alongside an auxiliary-loss-free load balancing strategy.
🔮 Future ImplicationsAI analysis grounded in cited sources
The AI industry will experience an intensified price war, leading to further commoditization of AI services.
DeepSeek's aggressive price cuts are designed to undercut major competitors, forcing other providers to re-evaluate their pricing strategies to remain competitive and potentially driving down costs across the board.
Increased adoption of DeepSeek's models will accelerate AI integration for cost-sensitive enterprises and developers.
The significantly lower API costs make advanced AI capabilities more accessible and affordable, enabling broader experimentation and deployment, especially for high-volume use cases and startups.
Geopolitical factors, particularly U.S. export controls on advanced chips, will continue to influence AI development and pricing strategies in China.
DeepSeek's optimization for Huawei's Ascend chips demonstrates a strategic pivot to domestic hardware, driven by restrictions on NVIDIA GPUs, which could foster a more self-reliant Chinese AI ecosystem and influence future pricing.
⏳ Timeline
2023-07
DeepSeek AI founded by Liang Wenfeng, backed by High-Flyer hedge fund.
2023-11
DeepSeek Coder, the company's first open-source code-focused model, is released.
2024-05
DeepSeek-V2, introducing multimodal capabilities, is launched.
2025-01
DeepSeek-R1 model and an eponymous chatbot are launched, with the mobile app briefly surpassing ChatGPT on the iOS App Store.
2026-03
DeepSeek V4, the newest flagship model, is launched, claiming 'era of cost-effective 1M context length'.
2026-05
DeepSeek announces permanent 75% price reductions for its flagship V4 Pro model.
📎 Sources (14)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 钛媒体 ↗