DeepSeek-V4-Flash Official API is now LIVE in public beta!

DeepSeek V4-Flash Official API is now LIVE in public beta! Deepseek v4 flash is chinese supremacy We've massive agent capabilities—benchmark scores are now far surpassing the V4-Pro-Preview. Check out the massive performance leap below! The official V4-Flash now natively supports the Responses API format and is fully adapted for Codex! Check out the official API docs: api-docs. Alrighty boys, Deepseek V4 Flash just dropped. This is from the same company if you remember two years ago when they initially released Deepseek, it was open source and crashed the United States tech market by like 30% one day. The entire stock market went down. When this original thing dropped because it was so smart and cheap and it basically showed the world that look how well China is doing with AI. And now two years later, it continues to do that. This is a tweet by Alex Finn, but you can host this model locally on less than, I think, 200 gigabytes of RAM. So you can do it on a 256 gigabyte Mac Studio, which I will be training soon. And again, it won't be a smart, it won't be as fast, it won't be as like intelligent, but you can do this. Not just that, look how much dumber it is than Opus 5. Only 18% dumber than Fable 5, only 17% dumber than Fable 5, only 12% dumber than K3. The intelligence factors already amazing, but look at the cost. So: ~15-18% less intelligent, ~99% cheaper. That's the whole pitch in one line. DeepSeek silently updated their changelog with a new V4-Flash upgrade 1 hour ago. Their new running v4 flash is massive +25.8 point leap from its initial April preview score of 56.9. Currently only available via their API, open weights release will follow shortly. On client tests, it showed that it is smarter than Fable 5, Sonnet 5 and almost as smart as GPT-5.6 Sol. I feel like I said this every day, but every single day it does happen. We're a Chinese model that's usually open weight or open source comes out and just cooks American models. And this is the most recent example with their new model, which isn't even new, it's a retrained old model. DeepSeek v4 flash is chinese supremacy