DeepCoder：Together 與 Agentica 推出達到 o3-mini 水準的 14B 完全開源程式碼推理模型
Original: [AINews] DeepCoder: A Fully Open-Source 14B Coder at O3-mini Level
After DeepSeek R1 set off a wave of open-source reasoning models, the open-source community saw many projects attempting to replicate its…
Together 與 Agentica Project（曾推出 DeepScaleR）聯合發表了全新 14B 程式碼推理模型「DeepCoder」。該模型完全開源，並採用 GPRO+（Group Relative Policy Optimization+）強化學習技術。在多項程式碼基準測試中，DeepCoder 展現出媲美 OpenAI o3-mini 的強大實力，是開源 AI 推理領域的重大突破。
After DeepSeek R1 set off a wave of open-source reasoning models, the open-source community saw many projects attempting to replicate its path to success. However, apart from Hugging Face's OpenR1, which continues to be actively updated, most projects have remained merely at the model distillation stage. Now, Together AI and the Agentica Project (the team behind DeepScaleR, which previously surpassed o1-preview with a 1.5B model) have joined forces to break this stalemate, officially launching a 14B-parameter fully open-source code reasoning model called **DeepCoder**.
Free shows the 3-line summary; Pro unlocks the full deep summary (~300 words) so you never have to click through.
See Pro plans →Want the original English / full article?
Read on TLDR AI (Buttondown) →Summaries are AI-generated; the original article is authoritative.