🔥36氪•Stalecollected in 29m
NetEase News and Bee integrate DeepSeek-V4 model
💡Major Chinese media adoption of DeepSeek-V4 signals the model's readiness for high-traffic production environments.
⚡ 30-Second TL;DR
What Changed
DeepSeek-V4 integrated into NetEase News and Bee
Why It Matters
This adoption demonstrates the growing enterprise-level deployment of DeepSeek models in Chinese media ecosystems. It highlights a shift toward using high-performance open-weight models for personalized content delivery.
What To Do Next
Evaluate DeepSeek-V4's API performance for your own content recommendation or generation pipelines to compare against proprietary models.
Who should care:Developers & AI Engineers
Key Points
- •DeepSeek-V4 integrated into NetEase News and Bee
- •Features include AI-assisted content creation and smart recommendations
- •Upcoming 'Campus Lobster Raising Plan' to gamify AI social interaction
🧠 Deep Insight
Web-grounded analysis with 29 cited sources.
🔑 Enhanced Key Takeaways
- •DeepSeek-V4 was officially released on April 24, 2026, as an open-source model under the MIT License, making its weights available for download, fine-tuning, and self-hosting.
- •The DeepSeek-V4 model is available in two versions: Pro, with 1.6 trillion total parameters and 49 billion active parameters, and Flash, with 284 billion total parameters and 13 billion active parameters, both supporting a 1 million token context window.
- •DeepSeek-V4 utilizes a Mixture-of-Experts (MoE) architecture and a Hybrid Attention Architecture, combining Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA), which significantly improves long-context efficiency by reducing KV cache memory and inference FLOPs compared to its predecessor.
- •DeepSeek-V4's API pricing is substantially lower than leading closed-source models, with DeepSeek V4-Pro costing approximately $1.74 per million input tokens and $3.48 per million output tokens, making it a highly cost-effective solution for large-scale AI applications.
- •NetEase has a broader strategic focus on integrating AI across its diverse business segments, including enhancing game development with AI-driven NPCs and content creation, and improving educational tools through its Youdao subsidiary, demonstrating a significant commitment to R&D.
📊 Competitor Analysis▸ Show
| Feature/Metric | DeepSeek V4 Pro | DeepSeek V4 Flash | OpenAI GPT-5.5 | Anthropic Claude Opus 4.7 | Google Gemini 3.1 Pro |
|---|---|---|---|---|---|
| Total Parameters | 1.6 Trillion (MoE) | 284 Billion (MoE) | N/A (Closed-source) | N/A (Closed-source) | N/A (Closed-source) |
| Active Parameters | ~49 Billion per token | ~13 Billion per token | N/A | N/A | N/A |
| Context Window | 1 Million tokens | 1 Million tokens (256K for some deployments) | ~1 Million tokens | ~1 Million tokens | N/A (Large) |
| Input Price (per 1M tokens) | $1.74 (standard), $0.435 (OpenRouter), $0.03 (cache hit) | $0.14 (standard), $0.112 (OpenRouter), $0.03 (cache hit) | $5.00 | $5.00 | $2.00 |
| Output Price (per 1M tokens) | $3.48 (standard), $0.87 (OpenRouter) | $0.28 (standard), $0.224 (OpenRouter) | $30.00 | $25.00 | $12.00 |
| License | MIT License (Open-source weights) | MIT License (Open-source weights) | Closed-source | Closed-source | Closed-source |
| Agentic Coding Benchmarks | Open-source SOTA in Agentic Coding benchmarks (e.g., 77/100 in Kilo CLI test, 55.4% on SWE-Bench Pro) | 60/100 in Kilo CLI test | 82.7% on Agentic Coding (GPT-5.5) | 91/100 in Kilo CLI test (Opus 4.7), 69.4% on Agentic Coding | 68.5% on Agentic Coding |
| World Knowledge/Reasoning | Leads open models, trails Gemini 3.1 Pro | N/A | N/A | N/A | Leads DeepSeek V4 on world knowledge |
| Multimodal Support | Native multimodal (text, images, video, audio) | N/A | N/A | N/A | N/A |
🛠️ Technical Deep Dive
- Model Architecture: DeepSeek-V4 is a Mixture-of-Experts (MoE) language model series. It retains the DeepSeekMoE framework and Multi-Token Prediction (MTP) strategy from DeepSeek-V3.
- Hybrid Attention Architecture: A key innovation is the Hybrid Attention Architecture, which combines Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA). CSA compresses KV caches along the sequence dimension and applies DeepSeek Sparse Attention (DSA), while HCA applies heavier compression with dense attention. This dramatically improves long-context efficiency.
- Parameter Counts: DeepSeek-V4-Pro has 1.6 trillion total parameters with approximately 49 billion active parameters per token. DeepSeek-V4-Flash has 284 billion total parameters with about 13 billion active parameters per token.
- Context Window: Both V4-Pro and V4-Flash models support a 1 million token context window by default, enabling processing of extensive documents and codebases.
- Efficiency: At a 1 million token context, DeepSeek-V4-Pro requires only 27% of the single-token inference FLOPs and 10% of the KV cache compared to its predecessor, DeepSeek-V3.2. V4-Flash further reduces this to 10% FLOPs and 7% KV cache.
- Training Data: V4-Pro and V4-Flash were pre-trained on 33 trillion and 32 trillion tokens respectively, a significant increase from V3's 14.8 trillion, with an emphasis on long documents and agentic execution traces.
- Post-training Pipeline: The post-training pipeline involves two stages: independent cultivation of domain-specific experts via Supervised Fine-Tuning (SFT) and Reinforcement Learning with GRPO, followed by unified model consolidation via On-Policy Distillation (OPD) from more than ten domain-specialist teacher models.
- Reasoning Modes: The instruct models support three reasoning modes: Non-think (fast, no chain of thought), Think High (explicit reasoning in
<think>blocks), and Think Max (maximum reasoning effort with a dedicated system prompt).
🔮 Future ImplicationsAI analysis grounded in cited sources
The integration of DeepSeek-V4 will significantly enhance NetEase's competitive edge in content creation and user engagement.
DeepSeek-V4's advanced AI capabilities, particularly in content generation, personalized recommendations, and efficient long-context processing, will allow NetEase to offer more sophisticated and engaging experiences on its News and Bee platforms.
DeepSeek-V4's cost-effectiveness and open-source nature will accelerate the adoption of frontier-level AI models across various industries, particularly in regions with compute constraints.
The model's significantly lower API pricing and open-weight availability make advanced AI more accessible and economically viable for a broader range of developers and companies, fostering innovation and competition.
The 'Campus Lobster Raising Plan' indicates a growing trend towards gamified and interactive AI social experiences.
This initiative suggests that AI integration in social platforms will move beyond passive recommendations to actively engage users through novel, gamified interactions, potentially creating new forms of community and content generation.
⏳ Timeline
2023-07
DeepSeek AI founded
2023-11
DeepSeek Coder, the company's first open-source model for coding, released
2024-05
DeepSeek-V2 model released
2024-11
NetEase Bee social content app launched
2024-12
DeepSeek-V3 model released
2025-01
DeepSeek-R1 reasoning model and eponymous chatbot application launched
2026-04-24
DeepSeek-V4 Preview (Pro and Flash versions) officially released
2026-05-16
NetEase News and Bee platforms integrate DeepSeek-V4 model
📎 Sources (29)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
📰
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: 36氪 ↗