DeepSeek发布V4.1 Flash模型 将承接V4 Pro所有调用请求

This article outlines key, practical updates from AI developer DeepSeek regarding its new large language model (LLM), revised API pricing structure, and automatic traffic migration for legacy interfaces—information relevant to all users of the company’s LLM services.

1. Key timeline and operational guidance: DeepSeek officially launched its V4.1 Flash model on September 10, 2026. Starting at 12:00 PM Beijing time on September 14, all API requests sent to the V4 Pro endpoint will be automatically routed to the new model. Users do not need to manually adjust their interface configurations, and all usage will be billed under the new Flash tier pricing, allowing zero-friction access to the updated version. The company has distributed official notifications of these changes to all API users, and detailed questions can be answered via official documentation.

2. Real-world performance advantages of the new model: Internal and external multi-round testing shows V4.1 Flash outperforms the older V4 Pro across four key metrics: performance, cost efficiency, response speed, and total task completion time, delivering a faster, higher-quality experience for future calls.

3. Practical pricing guidance: The new pricing uses a time-of-use tiered structure. During off-peak hours, cached input tokens cost only RMB 0.02 per million, uncached input tokens cost RMB 1 per million, and output tokens cost RMB 4 per million. All rates double during peak hours. Regular users can reduce costs by scheduling usage for off-peak windows and optimizing prompts to maximize cache hits.

DeepSeek’s series of moves—rolling out an upgraded LLM, revising its pricing structure, and implementing zero-friction traffic migration—carries strong reference value for brands planning AI-related business initiatives and optimizing return on investment.

1. Capture AI cost optimization opportunities: The newly launched V4.1 Flash delivers better performance than the legacy V4 Pro, paired with a low-threshold time-of-use pricing model. When building proprietary AI service modules, brands can prioritize this high cost-performance model to balance output quality and input costs.

2. Leverage the new model to improve service experience: V4.1 Flash offers faster response times and shorter total task completion, which will reduce user wait times for AI-powered brand services and improve the smoothness of user-brand interactions after integration.

3. Prepare for rule changes: After the automatic traffic migration on September 14, billing standards will shift. Brands’ technical integration teams should reconcile API call bills in a timely manner, schedule non-real-time batch processing tasks for off-peak hours, and reduce overall AI tool expenditure.

DeepSeek’s model iteration and pricing adjustment present clear cost-reduction and efficiency-improvement opportunities for sellers using AI tools to streamline operations, while also creating rule change risks that require advance preparation.

1. Access model upgrade benefits with zero friction: Starting at 12:00 PM on September 14, requests to the original V4 Pro endpoint will be automatically routed to V4.1 Flash. Sellers will gain direct access to the more powerful, faster-responding new model with shorter task processing times without manual interface adjustments, boosting operational efficiency across AI-assisted business scenarios.

2. Leverage pricing rules to cut operating costs: The new time-of-use tiered pricing offers lower rates during off-peak hours, with exceptionally low costs for cached requests. Sellers can schedule non-urgent batch processing tasks for off-peak windows, and maximize reuse of common prompt content to trigger cache hits and reduce AI tool spending.

3. Guard against cost overrun risks: All API rates double during peak hours, and post-migration billing will no longer follow original V4 Pro standards. Sellers should adjust calling strategies in advance, calibrate AI tool cost budgets, and avoid unnecessary peak-hour calls that would drive up operating costs.

DeepSeek’s launch of the higher cost-performance V4.1 Flash LLM and rollout of time-of-use pricing provide a new feasible path for factories advancing digital transformation and implementing e-commerce-related AI applications, while also unlocking related efficiency gains.

1. Lower cost threshold for digital transformation: The new model outperforms the earlier V4 Pro version, paired with tiered low pricing standards. When rolling out AI applications for production and operations, factories can adopt this high cost-performance model without bearing prohibitively high AI investment costs.

2. Model capabilities better aligned with operational scenario needs: V4.1 Flash’s faster response speeds and shorter total task completion times better match the rapid response requirements of production and operational links, reducing wait times across related systems.

3. Optimize AI usage strategies to cut investment: In line with the time-of-use pricing rules, factories can schedule non-real-time, high-compute tasks for off-peak processing. Teams should also note the September 14 automatic migration deadline, and reconcile calling cost changes across relevant systems in a timely manner.

DeepSeek’s iteration of the V4.1 Flash model and launch of its new time-of-use pricing system reflect the latest development trends in the LLM industry, while also providing a new technology selection direction for service providers to optimize service solutions and address client pain points.

1. Clear, referenceable industry development trends: Current LLM products are iterating toward stronger performance, lower costs, and faster response speeds, with billing models shifting from early uniform pricing to more flexible time-of-use tiered pricing. Service providers need to track these technology iteration trends closely and adjust their underlying technology selections to maintain service competitiveness.

2. New solutions for longstanding client pain points: Many clients face common AI service pain points including high costs, slow responses, and long wait times for extended tasks. The new model outperforms its predecessor across four core dimensions, with extremely low off-peak pricing. Service providers can build high cost-performance AI services based on this model, processing clients’ non-urgent batch demand during off-peak windows to help clients reduce usage costs.

3. Prepare for technical adaptation in advance: Complete interface compatibility testing ahead of the September 14 automatic migration to ensure service stability. Teams can also design tiered service offerings around time-of-use pricing to match different clients’ cost and timeliness requirements.

DeepSeek’s series of operations—launching the new model, implementing zero-friction automatic traffic migration, and rolling out time-of-use tiered pricing—carries strong reference value for the operational management, service optimization, and risk control of AI service platforms, while also reflecting core user demand directions for such platforms.

1. Referenceable ideas for user service optimization: Core user demand for AI services centers on four dimensions: stronger performance, lower costs, faster responses, and simpler operations. DeepSeek’s zero-friction automatic migration allows users to access the upgraded model without manual interface adjustments, drastically reducing user switching costs. This user-threshold-centered operational approach is worth emulating for platforms.

2. Referenceable direction for pricing rule design: Platforms can draw on the time-of-use tiered pricing framework to adjust peak and off-peak computing resource distribution via price levers, both improving computing resource utilization efficiency and providing lower-threshold access options for cost-sensitive users.

3. Key risk control points to note: For rule changes involving billing adjustments and interface routing, platforms should send clear, advance notifications to all users, ensure system stability during the migration window, respond promptly to user questions about bills and usage experience, avoid user dissatisfaction triggered by rule changes, and release supporting new model usage guides to guide users toward off-peak usage for lower costs.

DeepSeek’s series of moves—launching the V4.1 Flash model, implementing zero-friction traffic migration, and adopting time-of-use tiered pricing—reflect the latest trends in LLM industry development, providing a vivid real-world case for related industry research and business model analysis.

1. New trends in industrial technology iteration: The newly released V4.1 Flash outperforms the previous-generation V4 Pro across four dimensions: performance, cost efficiency, response speed, and total task completion time. This indicates that LLM technology iteration remains ongoing, inference-side costs still have significant room to decline, and industry competition has entered a new phase that prioritizes both performance and cost.

2. New features in business model innovation: LLM commercialization is shifting from early, extensive fixed-unit-price billing to refined time-of-use tiered pricing, using price mechanisms to optimize computing resource allocation—a key marker of maturing LLM commercialization. The real-world operational effects of this pricing model merit long-term tracking and research.

3. New shifts in vendor operational strategy: Using automatic routing migration to handle all legacy version requests both reduces user switching costs and accelerates large-scale adoption of the new version. This product iteration and market penetration strategy provides a new observation sample for research on LLM vendors’ user operation and go-to-market approaches.