DeepSeek has just dropped an upgraded translation of its already telling V3 model — and it ’s got developers talking . This Chinese AI startup released theV3andR1models originally this yr , and they straightaway grab attention by offering carrying out that rivals top - tier models from OpenAI and Google — entirely open - source and free .
Now , they are back at it again with the updated version of the V3 model – DeepSeek - V3 - 0324 . This is already father buzz for writing hundreds of line of code without breaking a sweat .
rent ’s break it down .
Table of Contents
What’s New in DeepSeek-V3-0324?
The big modification here is power . The parameter numeration jumped from671 billion to 685 billion , pass on it more mental ability while still using the efficientMixture - of - Experts ( MoE)architecture . Only 37 billion parameter set off per task , so it ’s smart with how it use imagination .
They also shift to theMIT licence , which is developer - friendly and makes integrating much well-heeled .
Benchmarks also show strong gain :
This is n’t just benchmark bagatelle , either . Here are the change that you will notice when using the new exemplar .
What You’ll Notice When Using It
Then How It Performs?
People have tested it — and the result are impressive .
Petri Kuittinen , a Finnish reader , got it to generate a fully responsive landing page for an AI company—958 lines of ferment code . Jasper Zhang , a Math Olympiad gold medalist , gave it a 2025 AIME job . It solved it cleanly .
Apple’sAwni Hannunran it on a512 GB M3 Ultra Mac . The fastness was around 20 + tokens per second , but the peak memory usage was just381 GB , which is substantial for a model this size .
We test it too .
When we asked it to produce a Python vane app using Flask , include login functionality and hash password security , it generate the code . To my surprise , it knead , too .
We essay the same on ChatGPT and Gemini . ChatGPT preserve restarting the yield . Gemini managed to finish it after a few tries , but the computer code was uncomplete and did n’t work without serious fixing .
How to Access the Latest DeepSeek V3?
you may directly get at the V3 from theDeepSeek websiteand the mobile app . By default option , it uses the new DeepSeek - V3 - 0324 model . So you could just hop on and try the fresh model right aside .
developer can integrate DeepSeek into their program and websites by using the API , which cost the same . you may use the sameAPI endpoint(model = deepseek - Old World chat )
To download and range the model topically , you’re able to do it from theHuggingFace chopine .
What’s Next?
Rumors sharpen to an upcoming R2 abstract thought model — perhaps even preferably than expected . And base on how ripe V3 - 0324 is , R2 could make an even braggart splash .
However , not everyone ’s thrilled . With its rising influence , DeepSeek is under U.S. government scrutiny over home security measures and data point secrecy . There ’s talking of banning its apps from prescribed machine . Still , DeepSeek - V3 - 0324 is proving that undecided - author AI can be powerful , practical , and price - effective . If you ’re a coder , constructor , or just rummy about what ’s next in AI , you should test it for yourself .
