Claude Code, who won the title of Deepseek, gave me such a Mediterranean hairstyle. Isn’t it handsome?
! It’s on par with Gemini. Both of them were beaten by GPT 6, but CC+DS has not started to optimize, and DS cannot generate textures, so it is still in a relatively primitive state. I'm wondering, if we use local 27B multi-modality, will it be better than DS, a model that cannot recognize images?
First impressions of DeepSeek V4.1 Beta.
This is one of my new benchmarks. I hope you like it. For a flash model and no vision this is impressive. Cost - $0.41 Total Token Use - 13 956 794
Deepseek dropped V4.1 Flash as a two-day test drive and the model ID says it all
Deepseek-v4.1-flash-expires-on-0910 new architecture, native multimodal, faster inference, same pricing as V4 Flash. 20 concurrent requests per account and then it's gone, come september 10th
OpenDesign Harness Demo 2, built with DeepSeek V4 Flash.
Your brief, your design system, your brand across every page. From typography to imagery, every detail helps tell your product’s story. See the transformation 👇
You have reached the end of the archive
All of deepseek