DeepSeek Launches Internal Testing of New Model; Users Report Blazing Speed
wallstreetcnOn Sept. 8, DeepSeek announced that an intermediate version of DeepSeek V4.1 Flash has entered internal testing. The model features a new architecture, natively supports multimodal capabilities, is more powerful, faster, and more cost-effective.
Developers can access it by keeping the base_url unchanged and setting the model name to deepseek-v4.1-flash-expires-on-0910. Billing is currently the same as DeepSeek V4 Flash, with a rate limit of 20 concurrent requests per account.
Same as DeepSeek V4 Flash, the current internal testing version of DeepSeek V4.1 Flash is priced at 0.05 yuan per million tokens for input (cache hit), 1.5 yuan for input (cache miss), and 4.5 yuan for output during off-peak hours; during peak hours, the corresponding prices are 0.1 yuan per million tokens for input (cache hit), 3 yuan for input (cache miss), and 4.5 yuan for output.
Based on the model name, this internal test will automatically expire and go offline on Sept. 10.
Some developers on X have already shared their experiences, with the highest output speed reaching 507 tokens/s.
One developer asked the model to create HTML with an SVG drawing of a 2D animation of a pelican riding a bicycle, and DeepSeek V4.1 Flash achieved an output speed of 328 tokens/s.
Another developer also tried the pelican riding a bicycle test; the image below shows the result. He revealed that the average output speed exceeded 300 tokens/s. Another user said that after manually adding the model, they found the speed to be very, very fast.
The DeepSeek model series has been updated frequently recently.
On July 31, the official version of DeepSeek V4 Flash API was first launched for public beta; on Aug. 13, the official version of DeepSeek V4 Pro API was released; on the same day, the developer preview version (v0.1) of DeepSeek Harness entered public beta and was open-sourced simultaneously. On Aug. 19, DeepSeek Harness was iterated to version v0.1.0-rc.8.
Subsequently, on Aug. 21, DeepSeek V4 Flash Vision Exp was launched on the DeepSeek API platform and officially open-sourced on Aug. 31.
Now, with DeepSeek V4.1 Flash entering internal testing, it may signal the start of a new round of model update cycles.
This content is for informational and educational purposes only and does not constitute investment advice related to BTCC. BTCC makes every effort but cannot guarantee the truthfulness, accuracy, or originality of the content above.