Abstract:
Just now, DeepSeek announced,
DeepSeek V4.1 Flash
The intermediate version has started internal testing. The new model adoptsnew model structure
,Native support for multi-modality, stronger capabilities, faster speed and lower cost
. The developer keeps the base_url unchanged and sets the model name todeepseek-v4.1-flash-expires-on-0910
It can be called. The current billing is the same as DeepSeek V4 Flash. Each account islimited to 20 concurrency
.
Same as DeepSeek V4 Flash, the price of the current internal beta version of DeepSeek V4.1 Flash is 0.05 yuan per million tokens for input (cache hit), 1.5 yuan for input (cache miss), and 4.5 yuan for output during idle periods; the corresponding price during peak hours is 0.1 yuan per million tokens for input (cache hit), 3 yuan for input (cache miss), and 9 yuan for output.
According to the model name, this internal test
will automatically expire and be offline on September 10th
.Some developers on social platform .
A developer asked the model to create HTML, and the content was to draw a 2D animation of a pelican riding a bicycle using SVG. DeepSeek V4.1Flash completed the task in only 75.4 seconds, with an output speed of
328 tokens/s.
Another developer also experienced the test of Pelican riding a bicycle. The picture below is the generated effect. He revealed that the average output speed exceeded
300 tokens/s
.
Another netizen said that after manually adding the model, he found that it was very, very fast.
DeepSeek series models
Frequently updated in recent times
.On July 31st, the official API version of DeepSeek V4 Flash was launched for public testing; on August 13th, the official API version of DeepSeek V4 Pro was released; on the same day, the developer preview version (v0.1 version) of DeepSeek Harness was officially tested and open sourced simultaneously. On August 19, DeepSeek Harness was iterated to version v0.1.0-rc.8.
Following this, on August 21st, DeepSeek V4 Flash Vision Exp was launched on the DeepSeek API platform and officially open sourced on August 31st.
Now, DeepSeek V4.1 Flash has started internal testing, which may mean that it will start a new round of model update cycle.
Comments