DeepSeek Announces Price Reduction for Flash Series Models: Output Cost Reduced to 4 Yuan per Million Tokens

According to the announcement of DeepSeek's open platform, the official will implement new pricing starting at 12:00 Beijing Time on September 10, 2026, adjusting the prices of the Flash series models.

After the price adjustment, the cost is 0.02 yuan per 1 million tokens for input cache hit during off-peak hours, 1 yuan per 1 million tokens for input cache miss, and 4 yuan per 1 million tokens for output. During peak hours, the price of each item is twice the off-peak hour price.

All three pricing tiers have decreased, with a price reduction rate of 60% for cache hits

Currently, the Flash series includes two models: deepseek-v4-flash and deepseek-v4-flash-vision-exp, and the price change range for these models is consistent.

Specifically, the input (cache hit) price has dropped from 0.05 yuan to 0.02 yuan during off-peak hours, and from 0.10 yuan to 0.04 yuan during peak hours, a decrease of 60%. The input (cache miss) has dropped from 1.5 yuan to 1 yuan during off-peak hours, and from 3.0 yuan to 2 yuan during peak hours, a decrease of 33.33%. Output has dropped from 4.5 yuan to 4 yuan during off-peak hours, and from 9.0 yuan to 8 yuan during peak hours, a decrease of 11.11%.

DeepSeek Releases 150 Positions at Once: Harness Team Urgently Recruiting Engineers, Scale Expansion Forces Backend Rewrite

DeepSeek Harness lead Cui Tianyi is hiring ~150 roles, citing demand from new directions and systems. Known for small teams and high talent density, this is DeepSeek's second expansion since June. It focuses only on server-side development and Agent elastic computing, unlike the earlier plan to scale all departments.....

DeepSeek Recruiting 150 Senior Engineers, Focusing on Agent Engineering and DSec Infrastructure

On September 8th, DeepSeek launched a recruitment of approximately 150 engineers, focusing on backend development and agent elastic computing research. The company is particularly looking for senior backend engineers with 2 to 10 years of experience, and no AI research positions are available. This move follows the previous department-wide expansion in June, indicating a strategic shift from pure model research to agent engineering and infrastructure construction, covering areas such as large model research platforms, agent framework components, and DeepSeek API.

Tencent's New Generation Large Model Hy4 Will Be Released, Becoming the Core Engine of WorkBuddy

The internal test interface of Tencent's new generation large model Hy4 has been exposed, with its release imminent. The interface shows that Tencent continues its strategy of self-research and open source, offering three options: the expert-level large model Hy4, the newly upgraded Hy3 suitable for most scenarios, and DeepSeek, which focuses on reasoning and understanding. Hy4 is the focus of this upgrade.

DeepSeek Announces Price Reduction for Flash Series Models: Output Cost Reduced to 4 Yuan per Million Tokens

All three pricing tiers have decreased, with a price reduction rate of 60% for cache hits

Related Recommendations

DeepSeek Releases 150 Positions at Once: Harness Team Urgently Recruiting Engineers, Scale Expansion Forces Backend Rewrite

DeepSeek Recruiting 150 Senior Engineers, Focusing on Agent Engineering and DSec Infrastructure

DeepSeek's Revenue Reached 475 Million Yuan in the First 7 Months, Surging 9 Times, 50 Billion Yuan Valuation and Second Round of Financing Under Way

Tencent's New Generation Large Model Hy4 Will Be Released, Becoming the Core Engine of WorkBuddy

DeepSeek Launches Official Public Account for the Harness Team, General Agent Product to Be Released Soon