The fifth issue of Tabbit AI model popularity list is released, the statistical period is from September 7th to September 13th.
DeepSeek’s performance in this issue is particularly outstanding. DeepSeek-V4-Flash-Vision-Exp has topped the list for three consecutive weeks, and the newly launched DeepSeek-V4.1-Flash also ranked second. Overall, all DeepSeek V4 models rank among the top ten.
✦ DeepSeek-V4.1-Flash was newly launched and ranked second
On September 10, DeepSeek officially released the new multimodal model DeepSeek-V4.1-Flash, and we accessed it immediately.
As the smallest model in DeepSeek's new generation architecture series, it uses a 552B parameter MoE architecture and a new Causal-Encoder-Decoder structure. The input activates only 8B parameters and the output activates 16B. The inference cost is significantly lower than known models of the same size.
The model natively supports multimodal visual understanding and is trained with larger-scale reinforcement learning. According to official benchmark test results, it has surpassed flagship models including DeepSeek V4 Pro in many evaluations, and the generation speed has also been greatly improved. In addition, the new generation model greatly compresses the KV Cache, and the HBM requirement is reduced to 1/4 of the previous generation, further reducing the inference cost of Agent-type tasks.
✦ DeepSeek-V4-Flash-Vision-Exp topped the list for three consecutive weeks
Since reaching the top in the third issue, DeepSeek-V4-Flash-Vision-Exp has topped the domestic list for three consecutive weeks, and the popularity of DeepSeek's multimodal models continues unabated.
✦ IMPORTANT NOTE
It should be noted that DeepSeek has announced that V4-Flash-Vision-Exp will be routed to V4.1-Flash on the day V4.1-Flash was launched. Therefore, we will take V4-Flash-Vision-Exp offline in the near future and recommend users to switch directly to DeepSeek-V4.1-Flash for conversation or task processing.
The above is the interpretation of the trend of the AI model popularity list last week. For those who have not tried Tabbit yet, you can visit the official website www.tabbit.com to download the desktop app for free and try it out.
To experience Tabbit on mobile, you can search for "Tabbit" in various app stores to download. By logging in to the same account, Campus Skills and data can be synchronized in the cloud.