Home
DeepSeek Unifies Three Modes: V4.1 Flash Launches Today at Fastest Speed, V4 Pro Will Be Discontinued
DeepSeek has consolidated its fast, expert, and image recognition modes into a single unified system, eliminating the need for manual switching.
This new intelligent mode automatically assesses query complexity, activates visual processing for image inputs, and scales computational resources based on task difficulty. Previously, the multimodal model DeepSeek-V4-Flash-Vision-Exp was available on the API platform, supporting Chat Completions, Messages, and Responses formats.

V4.1 Flash Surpasses V4 Pro; Legacy Models Scheduled for Retirement on September 14
DeepSeek announced the official launch of the V4.1 Flash model around September 10, 2026 (Beijing Time). Following extensive internal and external testing, V4.1 Flash demonstrates superior performance, cost-efficiency, and speed compared to V4 Pro.
The V4 Pro service will be discontinued at 12:00 Beijing Time on September 14, 2026. Post-discontinuation, traffic will be automatically routed to V4.1 Flash, with billing adjusted to reflect V4.1 Flash pricing.
According to an official notice, the intermediate version of V4.1 Flash utilizes a new architecture during testing, offering native multimodal support, enhanced capabilities, faster processing, and reduced costs.
Flash Series Pricing Update
Additionally, DeepSeek announced pricing adjustments for the Flash series effective at 12:00 on September 10. Off-peak input cache hit rates are priced at 0.02 yuan, while cache misses cost 1 yuan. Output pricing is set at 4 yuan, with peak-hour rates doubling the off-peak prices.
Related article
Deezer Reports Over 50% of Daily Uploads Are AI Music, Plans to Remove Unplayed Content After Six Months
Deezer, a major music streaming service, recently revealed alarming statistics: AI-generated tracks now make up over half of its daily uploads. In June, the platform recorded a surge in AI music submissions, averaging roughly 90,000 tracks per day.Th
How to optimize SEO for Google Japan and Yahoo Japan?
How to Build a Python Random Name GeneratorIntroductionSetting Up Vowel and Consonant ArraysCreating the Name Generation LogicBatch-Generating Random NamesExporting Results to CSVAdvantages and LimitationsConclusionFrequently Asked QuestionsWhether y
SpaceX Valuation Surges to $2.6T, Briefly Overtaking Amazon
SpaceX briefly surpassed Amazon to become the world’s fifth most valuable company, nearly overtaking Microsoft, before its shares retreated from those highs before the market closed on Tuesday.Following a 20% jump on its first full trading day, Space
Related Special Topic Recommendations
Comments (0)
0/500
DeepSeek has consolidated its fast, expert, and image recognition modes into a single unified system, eliminating the need for manual switching.
This new intelligent mode automatically assesses query complexity, activates visual processing for image inputs, and scales computational resources based on task difficulty. Previously, the multimodal model DeepSeek-V4-Flash-Vision-Exp was available on the API platform, supporting Chat Completions, Messages, and Responses formats.

V4.1 Flash Surpasses V4 Pro; Legacy Models Scheduled for Retirement on September 14
DeepSeek announced the official launch of the V4.1 Flash model around September 10, 2026 (Beijing Time). Following extensive internal and external testing, V4.1 Flash demonstrates superior performance, cost-efficiency, and speed compared to V4 Pro.
The V4 Pro service will be discontinued at 12:00 Beijing Time on September 14, 2026. Post-discontinuation, traffic will be automatically routed to V4.1 Flash, with billing adjusted to reflect V4.1 Flash pricing.
According to an official notice, the intermediate version of V4.1 Flash utilizes a new architecture during testing, offering native multimodal support, enhanced capabilities, faster processing, and reduced costs.
Flash Series Pricing Update
Additionally, DeepSeek announced pricing adjustments for the Flash series effective at 12:00 on September 10. Off-peak input cache hit rates are priced at 0.02 yuan, while cache misses cost 1 yuan. Output pricing is set at 4 yuan, with peak-hour rates doubling the off-peak prices.
Deezer Reports Over 50% of Daily Uploads Are AI Music, Plans to Remove Unplayed Content After Six Months
Deezer, a major music streaming service, recently revealed alarming statistics: AI-generated tracks now make up over half of its daily uploads. In June, the platform recorded a surge in AI music submissions, averaging roughly 90,000 tracks per day.Th
How to optimize SEO for Google Japan and Yahoo Japan?
How to Build a Python Random Name GeneratorIntroductionSetting Up Vowel and Consonant ArraysCreating the Name Generation LogicBatch-Generating Random NamesExporting Results to CSVAdvantages and LimitationsConclusionFrequently Asked QuestionsWhether y
SpaceX Valuation Surges to $2.6T, Briefly Overtaking Amazon
SpaceX briefly surpassed Amazon to become the world’s fifth most valuable company, nearly overtaking Microsoft, before its shares retreated from those highs before the market closed on Tuesday.Following a 20% jump on its first full trading day, Space











