← Back to live feed1 story
The updated architecture provides faster processing speeds and reduced costs for developers currently in the testing group. This intermediate version, known as V4.1 Flash, is now in internal beta and requires API users to set the model name to deepseek-v4.1-flash-expires-on-0910. The release introduces native multimodal support while maintaining the existing base URL for all requests.
Pricing for the trial version remains identical to the current deepseek-v4-flash. DeepSeek has limited access to 20 concurrent requests per account for the duration of the internal test.
Sign in to suggest edits
Key sources
- SOURCE@maxforai“每账号限流 20 并发”x.com
- SUPPORT@gpuhell“new model architecture with native multimodal support”x.com
- SUPPORT@jiqizhixin“Pricing remains the same as deepseek-v4-flash”x.com
- SUPPORT@kimmonismus“Official release post probably very soon”x.com
- SUPPORT@zoldener“intermediate build with: new architecture + native multimodality”x.com
- SUPPORT@plotarmordev“some users hit errors around 1700 requests with vram climbing”x.com
- SOURCEhuggingnewshuggingnews.com
- SOURCE@zephyr_z9“V4.1 Pro is also coming”x.com