DeepSeek V4 released: millions of contexts and open source, a new node in the domestic AI camp

Core conclusion

On April 24, 2026, DeepSeek quietly released a preview version of a new series of models, DeepSeek V4, without any pre-heating, which was simultaneously open sourced. This model has returned after nearly 16 months of silence. There are three core highlights: million-word ultra-long context, first release of Huawei Ascend chips, and API prices hitting the floor. For AI practitioners, this means that the cost of domestic inference has reached a new low, Agent applications have a looser context budget, and the Hugging Face ecosystem has an open source option worthy of attention.

Key Points

• Release time: April 24, 2026 (Beijing time)
• Model model: DeepSeek-V4-Pro (high-performance version) and DeepSeek-V4-Flash (lightweight version)
• Core Upgrade: Million-word super-long context, greatly enhanced Agent capabilities, 100% open source
• Adaptation dynamics: Huawei Ascend 910C is released for the first time, NVIDIA publishes a document on rapid adaptation, Baidu Qianfan Day0 access
• Pricing Strategy: V4-Pro model API is 25% off for a limited time until May 5, 2026

Background and trigger events

The release of DeepSeek V4 is a landmark event in itself. Nearly 16 months have passed since the last V3/R1 release. During this period, the global AI landscape has undergone significant changes: OpenAI iterated to GPT-5.5, Anthropic launched the Claude 4 series, Zhipu and MiniMax were launched successively, and Doubao and Qwen frequently topped the list.

During the quiet period, the outside world continued to speculate on whether DeepSeek would be unable to succeed. On April 24, an unannounced release finally gave the answer.

One detail worth noting: DeepSeek frankly wrote in the technical report that V4’s capability level still lags behind GPT-5.4 and Gemini-3.1-Pro, stating that “the development trajectory lags behind the cutting-edge closed-source model by approximately 3 to 6 months.” This proactive attitude of acknowledging the gap is quite rare among Chinese AI manufacturers, and it also enhances the credibility of technical reports.

Key Impact (by Dimension)

Adaptation suggestions

For readers of waytoclawearn (AI automation enthusiasts and money-making practitioners), DeepSeek V4 is one of the most noteworthy updates in recent times:

• API callers: Register a DeepSeek developer account now, take advantage of the 25% discount window (until May 5) to conduct large-volume testing, and reduce the cost of AI-driven content production pipelines
• Agent Developer: Millions of contexts mean that the Agent can "remember" the entire code base or hours of conversation records, and is suitable for building long process automation using orchestration tools such as n8n and LangGraph.
• Local Deployers: Follow the open source weight download of DeepSeek-V4 on Hugging Face to evaluate full replacement of legacy models in your personal workflow
• Content Creator: Use V4’s Agent capabilities and long context to process large documents to assist in batch content generation and SEO optimization

Task List (Reference)

• Register a DeepSeek developer account and obtain API Key
• Add V4-Pro as the inference backend to n8n or self-built pipeline
• Test a "whole book analysis" or "complete code base refactoring" scenario with millions of contexts
• Pay attention to the official pricing strategy after the discount ends on May 5

Example: DeepSeek V4 API call

# DeepSeek V4-Pro API（）
curl -sS -X POST "https://api.deepseek.com/v1/chat/completions" \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-pro",
    "messages": [
 {"role": "system", "content": " AI ，。"},
 {"role": "user", "content": " n8n + DeepSeek ， 500 。"}
    ],
    "max_tokens": 4096
  }'

• DeepSeek
• DeepSeek V4，
• DeepSeek-V4-Pro API 2.5
• Hugging Face: DeepSeek-V4

，tool_mentions OpenAI、GPT-5.5、Claude、Anthropic、DeepSeek、Qwen、n8n、LangGraph、Hugging Face、Gemini

Internal link guidance

• Want to learn how to use AI to automate content production? Watch: AI
• Real case: AI 3 ， 2 +
• Recommended tool: DeepSeek API