Open-source release Get DeepSeek-V4.1-Flash-1017: 🤗 Hugging Face: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017 🔗 ModelScope: https://modelscope.cn/models/deepseek-ai/DeepSeek-V4.1-Flash-1017

Tessera-27B-B-CJK is a Qwen3.8-27B coding agent that thinks in compressed CJK, and it spends fewer tokens doing it. Open weights, Apache-2.0. This is our first full "Tessera" lineup model drop! - Hugging Face: https://huggingface.co/ZelligeAI/Tessera-27B-B-CJK - Paper: https://zellige.ai/research/tessera-27b-b-cjk Tessera is our lineup of models and tunes that try new ways to save tokens, run more efficiently and show other parts of our research. What it does: Agent loops pay for thinking on every turn. So we taught Qwen3.8-27B to think less, and to think in compressed CJK, a shorthand that mixes English, Chinese words and symbols. The thinking and the output are kept apart. The model thinks in the shorthand, but its answers and tool calls stay in plain English and code, so your harness se

🔓 Open-source release Get DeepSeek-V4.1-Flash-1017: 🤗 Hugging Face: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017 🔗 ModelScope: https://modelscope.cn/models/deepseek-ai/DeepSeek-V4.1-Flash-1017

🚀 DeepSeek’s new model aims to build entire projects from scratch DeepSeek has released DeepSeek-V4.1-Flash-1017, an updated model designed to deliver stronger benchmark results, faster task completion, and shorter outputs than its predecessor. The goal is a better balance between performance and cost, with the model now available through DeepSeek’s website, app, and API. What makes this release interesting is its focus on building complete projects. Using reinforcement learning and self-feedback, the model is designed to improve its ability to develop projects end to end, from creating games to building 3D models. It’s also optimized for tools such as DeepSeekHarness, Tencent CodeBuddy, and WorkBuddy. DeepSeek recommends upgrading to Harness v0.2.1-rc.1 or later for the best experience. Source. @aipost 🏴