Runxin Xu

🔥DeepSeek-V4-Pro API is 75% OFF until May 5th, 2026, 15:59 (UTC Time)! Don't miss out on this massive discount. 🛠️Integration Updates: 🔹Claude Code: Set model to deepseek-v4-pro[1m] to unlock 1M context! 🔹OpenCode: Update to v1.14.24+ 🔹OpenClaw: Update to v2026.4.24+ Check the latest official API docs for full details: api-docs.deepseek.com/quick_start/pr…

We keep striving to build things that bring long-term value to everyone. We hope you enjoy our latest model — try it now on web, app, and API 🚀 「不诱于誉，不恐于诽，率道而行，端然正己」

@jiayi_pirate Good luck！

Today is my last day at xAI. It's been an intense, memorable year building in the frontier — I'm grateful to have worked with and learned from such talented and supportive colleagues. The journey itself is the reward. So long, and thanks for all the fish! 🐬

Introducing DeepSeek-V3.1: our first step toward the agent era! 🚀 🧠 Hybrid inference: Think & Non-Think — one model, two modes ⚡️ Faster thinking: DeepSeek-V3.1-Think reaches answers in less time vs. DeepSeek-R1-0528 🛠️ Stronger agent skills: Post-training boosts tool use and multi-step agent tasks Try it now — toggle Think/Non-Think via the "DeepThink" button: chat.deepseek.com 1/5

Ok, let's think deeper! 🧠✨

🚨BREAKING: DeepSeek-V3-0324🐳 just ranked #5 on the Arena leaderboard - surpassing DeepSeek-R1 and every other open model! Highlights: - #1 open model (MIT license) - 2x cheaper than DeepSeek-R1 - Top-5 across ALL categories - Significant jump over the previous DeepSeek-v3 Huge milestone for @deepseek_ai and a big win for open-source community! 👏

@ii_posts nice work，do you have scores in human last exam？

♊ Introducing II-Researcher: Our open-source AI research agent with strong benchmark performance ♊ 🚀 Key highlights: • New open-source framework for complex, multi-step search with reasoning tasks— a powerful, transparent alternative to proprietary systems • Two methods were implemented: finite state machines and tool-augmented thought processes with advanced reasoning models (R1 & QwQ) • Consistently delivers strong performance on GAIA and FRAMES benchmarks Come build with us and use II-Researcher!

🎯 Check out our DeepSeek-V3-0324! 🚀 We open-source this model, and hope it would be helpful to the community :)

Congrats! A comprehensive survey on Long Context Language Models👏!

🚀 One More Thing!

🚀 Day 0: Warming up for #OpenSourceWeek! We're a tiny team @deepseek_ai exploring AGI. Starting next week, we'll be open-sourcing 5 repos, sharing our small but sincere progress with full transparency. These humble building blocks in our online service have been documented, deployed and battle-tested in production. As part of the open-source community, we believe that every line shared becomes collective momentum that accelerates the journey. Daily unlocks are coming soon. No ivory towers - just pure garage-energy and community-driven innovation.

check out NSA for sparse attention 👏

Introducing CodeI/O (codei-o.github.io), a systematic way to condense diverse reasoning patterns via code input-output prediction to build massive training data for more reasoning tasks beyond commonly focused math problem-solving and code generation, which usually suffer from sparse and fragmented supervised signals. The idea is straightforward - teach the model to predict the expected output of a function given an input, or a feasible input given an existed output - but in a large scale with 3.5M instances! To do so, we transform large number of raw code files into executable functions in a unified format, and scale up the input-output pairs by simply executing them for multiple times. In training, we require the models to predict these inputs/outputs by pure natural language CoTs (instead of writing some code and call other interpreter to solve them) to promote the development of transferable reasoning skills that are independent of specific programming syntax. This simple code-derived task gives surprisingly robust performance gain on a broad range of tasks beyond code reasoning itself, including logic, symbolic, math & numerical, scientific, commonsense reasoning. ---------------------------------------------------- Another great advantage of CodeI/O is all data samples we constructed are verifiable (!), and we take full use of this feature to update our data to CodeI/O++ through multi-turn verification and revision, leading to better performance overall. We also hope these verifiable samples can boost RL training like DeepSeek-R1 beyond math and code generation as promising future work. ---------------------------------------------------- We have released our code, models and data to facilitate further research in the community. You can find more details by clicking these links: Website: codei-o.github.io Paper: huggingface.co/papers/2502.07… Code: github.com/hkust-nlp/Code… Released Resources: huggingface.co/collections/hk… ---------------------------------------------------- Finally, I would like to express my sincere gratitude to my collaborators @Guodaya, Dejian Yang, @pigjunebaba, my mentor @YuWu78356266, and my advisor @junxian_he for their help and for giving me the valuable opportunity to learn from them!

Today we’re making DeepSeek R1 and V3 available in Windsurf, making Cascade the first coding agent to support R1. It'll be half the cost to start with but we are committed to rapidly reduce this with time. R1 is truly fun and reading the chain of thought almost feels like a requirement for reasoning models.

Agreed. A learnable environment, and the model just want to learn. I have no doubt about this.

DeepSeek has not issued any cryptocurrency. Currently, there is only one official account on the Twitter platform. We will not contact anyone through other accounts.Please stay vigilant and guard against potential scams.

Every breakthrough reminds us there's still so much to learn and improve. Let's keep pushing boundaries together! #OpenSource #Innovation 🚀

Nice work! RL is truly fascinating😋!

@wzihanw 能不能帮哥们涨涨x的粉，不太会玩，哭了hh

@pigjunebaba Ju老师好！🤣

Enjoy our R1 😋

Nice job! Open research / open source accelerates progress.