CommunityNews

CommunityNews

DeepSeek-v3.2: Pushing the frontier of open large language models

We introduce DeepSeek-V3.2, a model that harmonizes high computational efficiency with superior reasoning and agent performance. The key technical breakthroughs of DeepSeek-V3.2 are as follows: (1) DeepSeek Sparse Attention (DSA): We introduce DSA, an efficient attention mechanism that substantially reduces computational complexity while preserving model performance in long-context scenarios. (2) Scalable Reinforcement Learning Framework: By implementing a robust reinforcement learning protocol and scaling post-training compute, DeepSeek-V3.2 performs comparably to GPT-5. Notably, our high-compute variant, DeepSeek-V3.2-Speciale, surpasses GPT-5 and exhibits reasoning proficiency on par with Gemini-3.0-Pro, achieving gold-medal performance in both the 2025 International Mathematical Olympiad (IMO) and the International Olympiad in Informatics (IOI). (3) Large-Scale Agentic Task Synthesis Pipeline: To integrate reasoning into tool-use scenarios, we developed a novel synthesis pipeline that systematically generates training data at scale. This methodology facilitates scalable agentic post-training, yielding substantial improvements in generalization and instruction-following robustness within complex, interactive environments.

Read in full here:

https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf

Where Next?

Popular Ai topics Top

AstonJ
Well done DeepMind… wonder what else they’re working on… One of biology’s biggest mysteries has been solved using artificial intelligen...
New
First poster: bot
In response to a national and international awakening on the issues of anti-Blackness and systemic discrimination, we have penned this pi...
New
CommunityNews
DeepMind’s New AI With a Memory Outperforms Algorithms 25 Times Its Size. DeepMind’s model, with just 7 billion parameters, outperformed...
New
First poster: CommunityNews
Getting a glimpse into Nvidia’s R&D has become a regular feature of the spring GTC conference with Bill Dally, chief scientist and se...
New
First poster: bot
AlphaTensor discovers better algorithms for matrix math, inspiring another improvement from afar.
New
New
First poster: mercyf
Google’s Veo 3 delivers AI videos of realistic people with sound and music. We put it to the test.
New
First poster: AstonJ
From fear to optimism: why I am convinced AI is worth embracing.
New
CommunityNews
Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing rout...
New
First poster: alvinkatojr
Open-source AI coding agent for your terminal. Engineered around DeepSeek’s prefix-cache. MCP first-class · plan mode · MIT.
New

Other popular topics Top

PragmaticBookshelf
Instantly view the changes you make to an app with stateful hot reload, and define a declarative UI in the same language as the app logic...
New
New
PragmaticBookshelf
Brace yourself for a fun challenge: build a photorealistic 3D renderer from scratch! In just a couple of weeks, build a ray tracer that r...
New
PragmaticBookshelf
Learn from the award-winning programming series that inspired the Elixir language, and go on a step-by-step journey through the most impo...
New
PragmaticBookshelf
From finance to artificial intelligence, genetic algorithms are a powerful tool with a wide array of applications. But you don't need an ...
New
AstonJ
I’ve been hearing quite a lot of comments relating to the sound of a keyboard, with one of the most desirable of these called ‘thock’, he...
New
AstonJ
Just done a fresh install of macOS Big Sur and on installing Erlang I am getting: asdf install erlang 23.1.2 Configure failed. checking ...
New
PragmaticBookshelf
Author Spotlight Mike Riley @mriley This month, we turn the spotlight on Mike Riley, author of Portable Python Projects. Mike’s book ...
New
PragmaticBookshelf
Build modern server-driven web applications using htmx. Whatever programming language you use, you’ll write less (and cleaner) code. ...
New
xiji2646-netizen
There’s a GitHub repo at forrestchang/andrej-karpathy-skills that’s sitting at 97.8k stars. It’s a single CLAUDE.md file with four behavi...
New