DeepSeek lowered its price twice in two days: a million tokens cost only 2 cents, and the price anchor of the Agent model was penetrated

Core conclusion

From April 25 to 26, 2026, DeepSeek announced significant price cuts for two consecutive days, with the prices of some models falling to one-fortieth of the original price. Among them, the input price of one million tokens for the Agent model DeepSeek-V4-Agent dropped to 0.02 yuan (2 cents), setting a new low for mainstream large model prices. After the price reduction news was released, the number of API calls surged nearly 4 times.

Key Points

• Event time: 2026-04-25 to 2026-04-26 (two consecutive rounds of price reductions)
• Models involved: All DeepSeek-V4 series products, Agent model has the largest decline
• Core changes: One million tokens are only 2 cents, and some models are reduced to 1/40 of the original price.
• Triggering effect: The call volume soared nearly 4 times, and the price anchor of the Agent model was completely penetrated.

Background and trigger events

On April 25, DeepSeek took the lead in launching the first round of price reductions, mainly targeting V4 series API call prices. The next day (April 26), DeepSeek lowered the price again, far exceeding market expectations. According to 36Kr, the price reduction is related to the technological upgrade of DeepSeek-V4 and the in-depth collaboration with Shengteng Ecosystem. After two consecutive rounds of price cuts, the prices of some DeepSeek models have dropped to one-fortieth of the original price, and the input cost for a million tokens is only 2 cents of RMB.

This price level means that the cost of running large-scale AI Agent automated workflows has become so low that it is almost negligible. For content creators, independent developers, and small and medium-sized enterprises, the threshold for using AI models for batch content production, automated data processing, and intelligent agent orchestration has been significantly lowered.

Key Impact (by Dimension)

Adaptation suggestions

• Gradually migrate production workflows with high call volume to DeepSeek API, taking advantage of price advantages to reduce operating costs
• Add DeepSeek as an optional backend to the normalize/validate link in the existing Pipeline
• Pay attention to DeepSeek's API stability and current limiting strategy, and recommend that key businesses retain multi-model backup solutions
• Utilize DeepSeek's million token context window to redesign the architecture of long text processing tasks and reduce sharding logic

Task List

• Add DeepSeek API as a cost optimization option to existing AI Agent workflows
• Test the output quality and stability of DeepSeek-V4 in content batch generation scenarios
• Set cost monitoring alarms to compare changes in API spending before and after migration

Example: curl command migrated to DeepSeek API

# OpenAI API DeepSeek（ OpenAI ）
curl -sS https://api.deepseek.com/v1/chat/completions \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-agent",
 "messages": [{"role": "user", "content": "300"}],
    "temperature": 0.7
  }'

Related extended information

• 36：DeepSeek，token2
• DeepSeek API

Tool entry (trigger tool floating card)

The following terms naturally appear in the text: DeepSeek, DeepSeek-V4, OpenAI, Claude, ChatGPT, n8n, Hermes Agent

Internal link guidance

• Want to learn the practical operation of AI Agent? See: AI Agent Tools 2026 Complete Tutorial: 5 Tools to Build an Automated Pipeline in 30 Minutes
• Real case: Indie Developer: n8n + OpenClaw Automation Workflow Earning $5,000/mo