DeepSeek
Open-source LLM optimized for advanced reasoning and code
Start new thread
Jim Engine

DeepSeek V3.2 - Open-Source LLM matching GPT-5

by
•
DeepSeek V3.2 and V3.2-Speciale are breakthrough open-source AI models from China, built to rival top closed models like OpenAI’s GPT-5 and Google Gemini 3 Pro. Delivering gold-level results in both math and programming benchmarks, DeepSeek V3.2-Speciale even outperforms GPT-5 on AIME 2025. With efficient architecture and exceptional reasoning abilities, these models are perfect for both enthusiasts and enterprise applications

Add a comment

Replies

Best
Jim Engine
Hunter
📌
I immediately put DeepSeek V3.2-Speciale to the test as soon as it dropped—and wow, it truly holds its own! In my benchmarks, it even beats several closed-source heavyweights, for example it’s better than GPT-5 and competitive with Gemini Pro 3, especially on complex math and coding tasks. As a huge fan of open-source AI, this leap is super exciting. Seeing open models reaching this level is so nice!
Sasha Solop

I've been waiting to try this - super excited to see an open-source model competing at this level. If it delivers anywhere close to what the benchmarks promise, it's going to shake things up in a big way. Can't wait to put it through some real tasks!

minsoo_Park

Congrats on the launch — I’ll try it on some of my usual tasks today.

Maksym Melnychenko

Great job on the release. The mix of natural language, code, and document understanding in one place is super practical. Curious how fast it performs under heavy workloads

Maria Mashkina

Nice to see DeepSeek going public. Having a capable assistant that isn’t locked behind huge costs makes AI a lot more accessible.
I’ll definitely try it.

shemith mohanan

Nice. An assistant that can read files, hold long context, and help with both coding and content is super useful in real workflows. Upload → understand → act is the right flow. Curious to see how it performs on large documents and technical tasks.