๐The Next Web (TNW)โขStalecollected in 32m
DeepSeek Makes 75% Price Cut Permanent, Escalating AI War
๐กDeepSeek's permanent 75% price cut is a major market shift that could significantly lower your AI infrastructure costs.
โก 30-Second TL;DR
What Changed
DeepSeek V4 Pro pricing is now permanently reduced by 75%.
Why It Matters
This move forces other AI providers to reconsider their pricing models to remain competitive. Developers and startups can expect lower operational costs for inference-heavy applications.
What To Do Next
Benchmark your current LLM inference costs against DeepSeek's new pricing to determine if migrating specific workloads could optimize your cloud spend.
Who should care:Founders & Product Leaders
Key Points
- โขDeepSeek V4 Pro pricing is now permanently reduced by 75%.
- โขNew token pricing ranges from $0.003625 to $0.87 per million tokens.
- โขThe move directly challenges major providers like OpenAI by significantly lowering the barrier to entry for high-performance models.
๐ง Deep Insight
Web-grounded analysis with 21 cited sources.
๐ Enhanced Key Takeaways
- โขThe permanent price cut positions DeepSeek V4 Pro's output tokens at $0.87 per million, making it approximately eight to nine times cheaper than OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.7 for comparable output token costs.
- โขDeepSeek V4 Pro is an open-weight Mixture-of-Experts (MoE) model featuring a 1 million token context window and the ability to output up to 384,000 tokens in a single request, matching or exceeding the specifications of several Western competitors.
- โขThis aggressive pricing strategy is consistent with DeepSeek's previous market entries, as the company similarly made a substantial promotional discount for its DeepSeek V3 model permanent at the end of 2024.
- โขDeepSeek V4 Pro demonstrates strong performance on agentic benchmarks, scoring around 91.2% on SWE-Bench Verified and 93.5% on LiveCodeBench, placing it in a similar performance tier to Claude Opus 4.7 and GPT-5.5 for complex coding and multi-step reasoning tasks.
๐ Competitor Analysisโธ Show
| Feature/Model | DeepSeek V4 Pro | OpenAI GPT-5.5 | Anthropic Claude Opus 4.7 | Google Gemini 3.1 Pro |
|---|---|---|---|---|
| Pricing (per 1M tokens) | ||||
| Input | $0.435 | $5.00 | $5.00 | $2.00 |
| Output | $0.87 | $30.00 | $25.00 | $12.00 |
| Context Window | 1M tokens | 1M tokens | 1M tokens | 2M tokens |
| Key Benchmarks | ||||
| SWE-Bench Verified | ~91.2% | Slightly above DeepSeek V4 Pro | ~93.9% | N/A |
| LiveCodeBench | 93.5% | N/A | 88.8% | N/A |
| CodeForces | #1 | N/A | N/A | N/A |
| Architecture | MoE (1.6T total, 49B active) | Proprietary | Proprietary | Proprietary |
| Open Weights | Yes (MIT License) | No | No | No |
๐ ๏ธ Technical Deep Dive
- Model Architecture: DeepSeek V4 Pro is a Mixture-of-Experts (MoE) model with 1.6 trillion total parameters and 49 billion activated parameters.
- Attention Mechanism: It incorporates a hybrid attention architecture that combines Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA) to enhance long-context efficiency. This design reduces single-token inference FLOPs by 27% and KV cache by 10% compared to DeepSeek-V3.2 at a 1M-token context.
- Connections & Optimization: The model utilizes Manifold-Constrained Hyper-Connections (mHC) to improve signal propagation stability across layers and employs the Muon optimizer for faster convergence and greater training stability.
- Post-training Pipeline: DeepSeek V4 Pro's post-training involves a two-stage pipeline: independent domain-expert cultivation (using SFT + GRPO) followed by unified model consolidation via on-policy distillation.
- Reasoning Modes: It supports three reasoning modes: Non-think (fast), Think High (logical analysis), and Think Max (full reasoning extent), allowing for flexible computational intensity based on task requirements.
- Output Capabilities: The model supports structured JSON output, as well as function and tool calling.
๐ฎ Future ImplicationsAI analysis grounded in cited sources
DeepSeek V4 Pro will significantly increase its market share in the LLM API space.
Its permanently reduced pricing, combined with competitive performance and a large context window, offers an unparalleled cost-efficiency that will attract a broad range of developers and enterprises.
The LLM price war will intensify, forcing major competitors to re-evaluate their pricing strategies.
DeepSeek's aggressive and sustained price cuts will exert immense pressure on companies like OpenAI, Anthropic, and Google to lower their own API costs to remain competitive and prevent customer migration.
There will be an accelerated shift towards cost-optimized AI model development and deployment.
DeepSeek's success in delivering high-performance models at a fraction of the cost demonstrates that efficiency is a critical differentiator, encouraging the industry to prioritize computational and memory optimization.
โณ Timeline
2016-02
High-Flyer, a hedge fund co-founded by Liang Wenfeng, is established, focusing on AI-driven trading.
2023-07
DeepSeek is founded as an independent AI lab and startup in Hangzhou, China, spun off from High-Flyer.
2024-05
DeepSeek-V2 is released, gaining popularity in China for its cost-efficiency and initiating a price war among Chinese AI companies.
2024-12
DeepSeek-V3 is released, with the company later making its promotional discount a permanent price.
2025-01
DeepSeek-R1 reasoning model and its mobile chatbot application are released, quickly becoming a top downloaded app and causing significant market impact.
2026-04
DeepSeek launches its V4 Pro and V4 Flash models, initially with a 75% promotional discount.
2026-05-24
DeepSeek officially makes the 75% discount on its V4 Pro model permanent, escalating the AI price war.
๐ Sources (21)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
๐ฐ
Weekly AI Recap
Read this week's curated digest of top AI events โ
๐Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: The Next Web (TNW) โ