DeepSeek launches v4.1 flash model
Archived — this story has rotated out of today’s deck. It is kept here in full.
The gist
DeepSeek launched V4.1 Flash beta on Sept 8, 2026, claiming it beats V4 Pro. The model natively handles text and images, pricing starts at $0.003 per cache hit.
Background
DeepSeek, a Chinese AI company, released an interim model called V4.1 Flash in a limited beta starting September 8, 2026. The model uses a new architecture that natively integrates multimodal capabilities, meaning it can process both text and images without external plugins. DeepSeek claims it outperforms its flagship V4 Pro on key metrics like performance, cost, and speed, while being cheaper. The beta is scheduled to expire on September 10, 2026, after which pricing adjustments will take effect.
How it unfolded
- Sep 8, 2026DeepSeek begins a limited-time beta of V4.1 Flash, accessible via API as 'deepseek-v4.1-flash-expires-on-0910'.
- Sep 9, 2026TechNode reports the beta, noting the model's new architecture and native multimodal support.
- Sep 10, 2026Beta is scheduled to go offline; pricing for Flash series adjusts at 12:00 Beijing Time, with off-peak rates at $0.003 for cache hits, $0.15 for cache misses, and $0.6 for output.
Who’s saying what
- Official
- DeepSeek states V4.1 Flash comprehensively surpasses V4 Pro across all key metrics and will route Pro requests to Flash at Flash's price until V4.1 Pro releases.
- Analysts
- Some analysts note the rapid iteration (40 days after V4) and question whether it should be called V5 Flash, while others see it as a strategic shift toward cheaper, capable models.
- Public
- A Hacker News user commented 'Waiting to use it', indicating anticipation among developers.
Still unverified
The claim that V4.1 Flash 'comprehensively surpassed V4 Pro' is based on DeepSeek's own testing; independent benchmarks are not yet available. The model is in beta and may have limitations, such as overthinking tendencies noted in a review.