Sources
Loading...
Additional media
Loading...
DeepNewz, mobile.
People-sourced. AI-powered. Unbiased News.
Together AI has launched DeepCoder-14B-Preview, a fully open-source code reasoning model that boasts an accuracy of 60.6%. This model, developed in collaboration with the Agentica team, is designed to rival the o3-Mini model while utilizing just 14 billion parameters. The release is part of a broader trend in artificial intelligence where companies are focusing on creating efficient and capable AI agents. OpenAI has also announced the open-sourcing of BrowseComp, a new benchmark for evaluating the web browsing capabilities of AI agents. These developments highlight the ongoing advancements in AI technology and the competitive landscape among AI developers.
A February 2025 research paper detailed CODEI/O, a novel approach from DeepSeek to make LLMs better at reasoning. #datascience #AI #artificialintelligence https://t.co/KJQMgSAkOh
DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning DeepSeek-AI et al.: https://t.co/qLA0cZyTkb #ArtificialIntelligence #DeepLearning #ReinforcementLearning https://t.co/VBFgMgFdfO
📝 DeepCoder: A Fully Open-Source 14B Coder at O3-mini Level DeepCoder-14B-Preview is a code reasoning model that achieves impressive performance compared to much larger models using reinforcement learning. The authors have open-sourced their code, dataset, logs, ... 1/7 https://t.co/HkRCg1YRWh
