DeepSeek just open sourced its high-performance, large-scale inference framework DeepInfer in Apache-2.0 license. github.com/deepseek-ai/DeepI… Supported features: - Optimized kernels, model and sampler runtime - Parallelism: Attention DP / balanced CP, MoE EP + EPLB - DSpark and confidence-scheduled verification - SWA bounded replay - Asynchronous scheduling (zero-overhead batch scheduler) - Batch-invariant inference - Deterministic sampling - Structured output

🚀 Introducing DeepInfer, our in-house inference framework, now fully open source! Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable. 🔗 github.com/deepseek-ai/DeepI… 1/4

为善与众行之，为巧与众能之

Have a try!

Always improving.😄

为充分发挥新模型 DeepSeek-V4.1-Flash-1017 的能力，请升级至 DeepSeek Harness 0.2.1-rc.1 及以上版本。

Make your wish come true 🚀🚀

Vision-related capabilities are rapidly advancing since we introduced vision to our flagship model. Now everyone can describe an idea and build their own games!

🚀 Today's DeepSeek Harness 0.2.1 release supports the new DeepSeek-V4.1-Flash-1017 model. The new model was specifically optimized for @DeepSeekHarness during training, improving task performance while shortening task runtime and reducing output usage and token costs. To take full advantage of the new model’s capabilities, please upgrade to DeepSeek Harness 0.2.1-rc.1 or later.

🚀 DeepSeek-V4.1-Flash-1017 is here! Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API. 🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. Upgrade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model. 1/4

Between 8:11 and 13:04 (Beijing Time) on October 15, 2026, DeepSeek API service system experienced a failure that prevented the API service from functioning properly during that period. We sincerely apologize for the inconvenience and disruption this may have caused. We immediately launched a comprehensive investigation and thorough rectification effort focused on service stability, and have comprehensively reviewed and eliminated single points of dependency to resolutely prevent similar issues from recurring. As compensation, we will fully refund the amount consumed within 7 days (10.8-10.14) on your account in the form of credit. The credit will be automatically deposited, and API billing deductions will be made from the credit account first. We apologize once again for the impact this failure had on you, and we thank you for your continued trust and support. 北京时间 2026 年 10 月 15 日 8:11 至 13:04 期间，我们的服务系统出现故障，导致该时段内 API 服务无法正常使用。对于此次故障给您带来的不便与困扰，我们深表歉意。我们已第一时间启动针对服务稳定性的全面排查和彻底整改，全面梳理并剥离单点依赖，坚决杜绝同类问题再次发生。 作为补偿，我们将按您账号 7 日内（10.8-10.14）的消耗金额以赠金形式全额返还，赠金自动到账，API 计费扣款先从赠金账户扣除。此次故障对您造成了影响，我们再次致歉，感谢您一直以来的信任与支持。

Test Gemini 4 Argon on a task from my hidden menu. Seems to be cooking.

Cool work about reusing previous model rollouts!

PhD defended today! 🎓 Huge thanks to my supervisor @WenhuChen for all the guidance and support throughout this journey, and to all my committee members for their time, thoughtful questions, and feedback. Next stop: joining Meta FAIR as a Research Scientist. Excited for what’s ahead!

Now math is doing math. What should humans do is highly puzzling

Three year ago from @ilyasut

How fast is AI's research taste improving? We find that the experimental research taste of frontier models has doubled every ~3 months since December 2025. The best model, Opus 5.5, now exceeds our expert human baseline. Our human experts are experienced researchers, but most haven’t worked at a frontier lab. Why measure research taste? In the AI Futures Model, it largely determines how quickly artificial superintelligence is reached once coding is fully automated.

Please check out @DeepSeekHarness. We just released packaged desktop versions for macOS and Windows; Linux users can get it from the @​deepseek-ai/dsh package on npm. deepseek.com/harness

Do you really need to finish every rollout in RL for LLM? Not if you make your critic better—and trust it! We propose Actor-Critic with Action Chunking (AC2): use a learned critic to score chunks of tokens => only need partial rollouts ⇒ train faster than GRPO!

DeepSeek Harness for desktop is now available on macOS and Windows.