DeepSeek keeps V4 Pro API alive after calling off shutdown

V4 Pro was due to go offline on September 14. On September 11, DeepSeek said it would keep serving API calls, with billing unchanged. The earlier plan had routed all V4 Pro requests to V4.1 Flash.

Rachel NwosuAI, models and technology

AI & modelsNewsRachel NwosuPublished: 11 September 20264 min readSources 3

DeepSeek keeps V4 Pro API alive after calling off shutdown

DeepSeek reversed a shutdown plan within three days. On September 10, the company released V4.1 Flash and told API users by email and through its open platform that V4 Pro would be retired at 12:00 Beijing time on September 14, 2026. From that point, V4 Pro requests would go to V4.1 Flash and be billed at V4.1 Flash rates.

The decision rested on testing. DeepSeek said internal and external tests showed V4.1 Flash had surpassed V4 Pro on performance, cost, speed and total time. The company concluded that moving requests wholesale to the new model would be better for users.

Then September 11 changed things. DeepSeek announced that, in response to user demand, it would keep DeepSeek V4 Pro API calls available after September 14, 2026, with billing unchanged. Teams still running V4 Pro in production can now migrate at their own pace.

Price differences are the most practical part of any migration decision. Under the published rates, deepseek-flash costs 0.02 yuan per million tokens for input cache hits during off-peak hours, 1 yuan for cache misses and 4 yuan for output. deepseek-v4-pro costs 0.15 yuan, 4.5 yuan and 13.5 yuan for the same three items. Prices double during peak hours.

For users, the episode is a reminder that model shutdowns and routing changes hit cost and stability directly. Keeping model names in one place, retaining a fallback, and running batch jobs off-peak are all preparations that can be made in advance. DeepSeek has not announced a launch date for V4.1 Pro.

The reversal also reflects the pace of iteration among Chinese large models. New releases come thick and fast, old models have shorter lives, and providers have to balance cost against stability while developers increasingly treat models as swappable components. DeepSeek handed the choice back to users, which lowers the urgency of migration in the short term. Once V4.1 Pro arrives, the picture may shift again.

Comments 0

Sources

All figures and quotations in this text come from the sources listed below.

Content prepared by the editorial team with AI assistance.

Rachel Nwosu

AI, models and technology

Rachel Nwosu covers AI, models and technology for FLASH24, working from public model documentation, benchmark releases and repository histories rather than press summaries, and she skips announcements that arrive without reproducible numbers. She checks training-data claims against dataset cards and reruns reported metrics where code is available. She spends much of her week interviewing researchers and engineers, tracking model launch calendars, and comparing vendor benchmarks with independent evaluations. Outside the desk she runs 3D printers, restores old computers, and tests how models learn from internet junk. She does not publish benchmark figures she cannot trace to a source.

Newsroom →

Comments

No comments yet — be the first.