Artificial Intelligence

Chinese tech company Baidu said on Sunday it has launched two new artificial intelligence (AI) models, which are said to offer the same performance as the Chinese chatbot DeepSeek but at half the cost.

• Research.

hpcaitech (the ColossalAI team) has officially released Open-Sora 2.0, an open-source video generation model with 11 billion parameters that has drawn widespread attention for balancing cost and performance. With only about $200,000 in training costs (equivalent to 224 GPUs), the model performs close to top commercial models in multiple evaluations.

Today we release OLMo 2 32B, the most capable and largest model in the OLMo 2 family, scaling up the OLMo 2 training recipe used for our 7B and 13B models released in November. It is trained up to 6T tokens and post-trained using Tulu 3.1. OLMo 2 32B is the first fully-open model (all data, code, weights, and details are freely available) to outperform GPT3.5-Turbo and GPT-4o mini on a suite of popular, multi-skill academic benchmarks. It is comparable to the leading open-weight models while requiring only a fraction of training compute. For example, OLMo 2 32B takes only one third of the cost of training Qwen 2.5 32B while reaching similar performance. The OLMo 2 family of models—now available in 7B, 13B, and 32B parameter sizes, all can be finetuned on a single H100 GPU node, and all models are available on the Ai2 playground.

Some of the world’s most advanced AI systems struggle to tell the time and work out dates on calendars, a study suggests.

Introducing Gemini Robotics, our Gemini 2.0-based model designed for robotics

Today Google releases Gemma 3, a new iteration of their Gemma family of models. The models range from 1B to 27B parameters, have a context window up to 128k tokens, can accept images and text, and support 140+ languages.

we’re launching a new set of APIs and tools specifically designed to simplify the development of agentic applications:

• The new Responses API⁠(opens in a new window), combining the simplicity of the Chat Completions API with the tool use capabilities of the Assistants API for building agents
• Built-in tools including web search⁠(opens in a new window), file search⁠(opens in a new window), and computer use⁠(opens in a new window)
• The new Agents SDK⁠(opens in a new window) to orchestrate single-agent and multi-agent workflows
• Integrated observability tools⁠(opens in a new window) to trace and inspect agent workflow execution

These new tools streamline core agent logic, orchestration, and interactions, making it significantly easier for developers to get started with building agents. Over the coming weeks and months, we plan to release additional tools and capabilities to further simplify and accelerate building agentic applications on our platform.

The enormous computing resources needed to train neural networks for artificial intelligence (AI) result in massive power consumption. Researchers at the Technical University of Munich (TUM) have developed a method that is 100 times faster and therefore much more energy efficient. Instead of taking an iterative approach, the parameters are computed directly based on probabilities. The results so far are comparable in quality to existing iterative methods.

The world's largest contract electronics maker, Foxconn, said Monday it has built its own large language model with reasoning capabilities, developed in-house and trained in just four weeks.

Large Language Models (LLMs) have demonstrated remarkable performance in solving complex reasoning tasks through mechanisms like Chain-of-Thought (CoT) prompting, which emphasizes verbose, step-by-step reasoning. However, humans typically employ a more efficient strategy: drafting concise intermediate thoughts that capture only essential information. In this work, we propose Chain of Draft (CoD), a novel paradigm inspired by human cognitive processes, where LLMs generate minimalistic yet informative intermediate reasoning outputs while solving tasks. By reducing verbosity and focusing on critical insights, CoD matches or surpasses CoT in accuracy while using as little as only 7.6% of the tokens, significantly reducing cost and latency across various reasoning tasks.