Product Thumbnail

DeepSeek-V3-0324

Code like 3.7 but open source

Open Source
Artificial Intelligence

DeepSeek-V3-0324 is the significant, open-source (MIT) update to DeepSeek's V3 model. It improves coding abilities nearing Claude 3.7 Sonnet, plus reasoning improvements.

Top comment

Hi everyone!

DeepSeek-V3-0324 is generating a lot of buzz in the community! It's a seemingly "minor" update (just a date code), but users are reporting major improvements, especially in coding.

From the community testing results:

💻 Coding Power: Multiple users are saying its coding abilities are now close to Claude 3.7 Sonnet. That's a huge jump.
🧠 Reasoning Improvements: It seems to have gained some of the reasoning capabilities of DeepSeek-R1, even though it's not explicitly a "reasoning model."
📝 Improved creative writing compared to previous DeepSeek models.
🔓 Still fully open-source with MIT license.

This update hasn't been officially announced in detail – typical DeepSeek style – but the early reports are very promising. It looks like DeepSeek is seriously leveling up its game.

Comment highlights

Tried accessing the demo, and the response speed seemed a bit inconsistent. Not sure if it's just temporaray, but worth checking.

Would be interesting to know where this one stands in comparison to models like GPT-4 and Claude.

Looking forward to seeing how well it performs across different types of queries.

This update sounds impressive! I'm excited to see how DeepSeek-V3-0324 enhances coding capabilities and reasoning skills. Could you share more about the specific improvements compared to the previous version? It would be great to understand the practical applications of these advancements!

I've tested DeepSeek V3 0324 last night, and it is good to generate interactive HTML report. However, 8k output is far from enough. I hope you guys can increase 8k output limitation to 32k or even more. I know that is a big challenge.

Haven’t tried it yet, but big applause to the DeepSeek team for their open-source contribution — it’s exciting to see this kind of work pushing the field forward!

I love how boldly @DeepSeek competes against the popular models. It's always something to look forward to when deepseek releases something. Congratulations on the launch!

i just admire the hard work DeepSeek team does. Is it already updated on their website though?

I tried it on a small Python script, and it nailed the suggestions almost perfectly, especially around error handling. Good job :)