Got deepseek V4.1 flash running locally on a 16GB m1 mac mini
Original FP4/FP8 weights, ssd streaming + custom mlx runner 108s ttft and about 23s/token (not to be confused with tok/s)
DeepSeek V4.1 flash is no longer hidden. Starting from 9/14
DeepSeek V4.1 flash is no longer hidden. Starting from 9/14, Pro file requests will be automatically routed to Flash and priced accordingly - which is equivalent to a price reduction across the board. I think Liang Sheng is back. After testing several cases, the speed is really amazing. Is this still the same DeepSeek that asked questions and waited for half a year for answers?
Deepseek V4.1 just created this Bank heist fps shooter game in one shot using our threejs game creation skill
Prompt : "Create a AAA Bank heist fps shooter game using three.js. Final output should be a single HTML file." Try it live : Skill in
Jundot/DeepSeek-V4.1-Flash-oQ3e-mtp
Designed by OMLX PR#3574 and Astra/implemented by Opus5 M3 Ultra 512GB over 260K (300GB over) Real time about 20tok/s Model release/quantization/PR/oMLX Thank you to everyone
You have reached the end of the archive
All of deepseek