DeepSeek V3 low-key update: big programming boost, R2 model coming too?

There's no launch party. There's no overwhelming publicity.March 24 DeepSeek Version V3-0324 is quietly online!.

https://huggingface.co/deepseek-ai/DeepSeek-V3-0324

This update is not part of the new generation of R2, but the enhancements are far more than expected, especially in the编程能力、前端UI design和上下文理解etc., have been able to match the Claude 3.7 Sonnet.

Another important upgrade is that DeepSeek continues to adhere to the spirit of open source by using directMIT Open Source Protocol, allowing developers the freedom to modify and commercialize.Completely breaks the monopoly of closed-source AI on top code models.

What does it all mean?

DeepSeek V3 itself is recognized as a cost-effective open source model in the industry, and this 0324 release, while only "Minor Version Updates", but it's a big boost. Yes, DeepSeek is that understated.

🔹 Parameter scale is raised again: the model parameters are increased from671 billion to 685 billion, although the increase is small, the optimized computational efficiency delivers aMore fluid reasoning skills.

🔹 Code ability skyrockets: The new version of DeepSeek V3 inFront-end development, website building, UI designThe aspect of the ability to usher in a significant improvement, and everyone basically agrees that itComparable to Claude 3.7 Sonnet.

🔹 Math reasoning is stronger: Multiple rounds of conversation and improved logical reasoning skills have resulted in more consistent performance on math problem solving, with some solutions approaching the reasoning model.

🔹 Open source licenses are more liberal: The adoption of the MIT license allows developers toModify, optimize and commercialize at will, dramatically lowering the barriers to AI adoption.

In tests with professional users, theDeepSeek V3-0324's code capabilities have reached the level of the world's strongest open source AI, even surpassing Claude 3.7 in some scenes.

Some of you have said that in the Aider benchmarks, theDeepSeek V3-0324 is significantly improved, approaching Claude Sonnet 3.7.

The example below is even more obvious and directly compares several top models. This webmaster has them generating code for animated weather cards.

The results are intuitive, and the new V3 is far superior to the original V3 and R1, and directly comparable to Claude 3.7.

One AI developer exclaimed:"No LLM on the market since June 2024 has come close to Claude 3.7 in terms of front-end code generation, but DeepSeek V3-0324 has finally done it."

The biggest "hidden egg" in this update is theDeepSeek Completely Unlocks MIT LicenseThe V3-0324 has been made to be aFully open, commercially available top AI models.

Why is the MIT license so important?

🔹 Unhindered commercial use: Businesses and individual developers are free to use it without paying expensive license fees.

🔹 Model free modification: It can be modified, fine-tuned and even used for your own products or projects without official restrictions.

🔹 Promoting the AI ecosystem: Compared to OpenAI's GPT-4.5 and Claude 3.7, thisclosed-source business model, DeepSeek makes it possible for more startup teams and individual developers toEqual access to top AI capabilities, true "technology parity".

A foreign netizen said, "DeepSeek is doing what Meta promised but failed to do, making AI really go open source and free."

But some people are happy and some people are sad.If open-source AI like DeepSeek continues to rage, what else can closed-source AI rely on to attract users?

The release of DeepSeek V3-0324 not only elevates the status of open source AI in programming and math, but also puts unprecedented pressure on closed-source AI companies like OpenAI and Anthropic.

Previously, closed-source AI was also possible with "greater capacity"Attracting high-end users, but nowadays, theFree and open source DeepSeek V3 is already approaching Claude 3.7 and even better in some areas.

Some overseas users have said it outright:"Seeing the advances in these open-source models makes me wish commercial AI companies would lower their prices."

This upgrade to DeepSeek has also caused people to start taking a fresh look at the future of the AI industry - theAs high-performance AI moves towards open source, will the business model for AI need to change radically?

So, here's the next question--Can R2 be far behind?

Many DeepSeek users have reported thatIf the V3-0324 is already this strong, then the R2 might actually topple the entire AI ecosystem.

The coming months may see even greater changes in the AI space, and DeepSeek, is already at the center of the storm.