AI News

Follow the latest AI and automation news

4 Aug 202621 Safar 1448 AH

[AINews] Qwen 3.8 Max(2.4T) and 27B, new open weights models for Coding and Cowork

Following doubts about Qwen's future in releasing new models, the lab has launched Qwen 3.8 Max, a massive 2.4T model. This model would have been the top open model globally if not for the Kimi K3 release. Qwen offers API services at $2 for input and $6 for output per million tokens. Qwen 3.8 Max features advanced capabilities such as long-horizon autonomous coding, having built a self-evolving coding harness over ten days. It also autonomously rebuilt a complete research pipeline, surpassing the original paper's benchmark by 2.71 points.

Agents

4 Aug 202621 Safar 1448 AH

Wukong Disappears Four Months After Launch: Alibaba Consolidates Into Qianwen Office Public Beta as Big Tech Races to Integrate AI Work Into Single Unified Agent Products

On August 3, Alibaba launched the public beta of Qianwen Office, an AI-based office platform. This system allows users to describe goals while the agent decomposes tasks and calls tools to deliver usable results, initially linked to DingTalk. The new platform integrates QoderWork, Wukong, and MuleRun, with the former Wukong unit upgraded to a first-tier division under the Alibaba Token Hub group. Four and a half months earlier, Wukong was central to Alibaba's AI strategy.

Models

4 Aug 202621 Safar 1448 AH

After Vowing to Make Pangu World No.1, Yu Chengdong Opens openPangu-2.0-Pro: 505B Parameters, 180B Sparse Activation, 512K Context, First Frontier Model Fully Trained on Ascend NPU

On July 31, Huawei launched openPangu-2.0-Pro as an open-source model featuring 505 billion parameters and 180 billion sparse activations, with a context length of 512K. This model is notable for being the first fully trained on non-NVIDIA hardware, utilizing 34 trillion tokens for training. Huawei provides a fully reproducible chain from training to inference, opening seven components including pretraining and post-training code. Yu Chengdong explained that the choice of 505 billion parameters was influenced by the limited allocation of computing power available to Huawei.

💡

What the DeepSeek Kill Line Actually Cuts: 8 Trillion Daily Tokens on a Single Tool, 85x Price Gap, and the Squeeze on Mid-Tier Models in Agent Economics

How might this price gap influence model selection in the market?

Models

SenseTime Open-Sources SenseNova U1.5-Lite-Preview: Native 4K Direct Output, 8B-MoT Lightweight Unified Multimodal Model With Precise Image Editing and Design Framework Replication

4 Aug 202621 Safar 1448 AH

Models

4 Aug 202621 Safar 1448 AH

SenseTime Open-Sources SenseNova U1.5-Lite-Preview: Native 4K Direct Output, 8B-MoT Lightweight Unified Multimodal Model With Precise Image Editing and Design Framework Replication

SenseTime has announced the launch of the SenseNova U1.5-Lite-Preview, a lightweight unified model that integrates language and visual semantics. This model features native 4K output and contains 8 billion parameters, enabling high-density image generation even when zoomed in. Operating under the NEO-Unify framework, the model can handle complex tasks such as design editing and content reorganization. In a notable test, it successfully edited a character in a maritime image, showcasing its understanding of design and creativity.

Business

New ways to learn and teach with ChatGPT Work and Codex

4 Aug 202621 Safar 1448 AH

Business

4 Aug 202621 Safar 1448 AH

New ways to learn and teach with ChatGPT Work and Codex

OpenAI has announced the launch of new educational plugins for ChatGPT Work and Codex, aimed at K-12 teachers and college students. These plugins enable educators to enhance their teaching methods and provide advanced research tools for students, significantly improving the learning experience. The plugins also feature unique capabilities such as creating customized educational content. This development is crucial in promoting the use of artificial intelligence in education, as it offers innovative resources that support self-directed learning and assist teachers in delivering more interactive and effective content.

Governance

Disrupting a Criminal Scam Operation

4 Aug 202621 Safar 1448 AH

Governance

4 Aug 202621 Safar 1448 AH

Disrupting a Criminal Scam Operation

OpenAI has uncovered a scam operation based in Cambodia that utilized ChatGPT to support fraudulent investments, romance schemes, gambling, and impersonation tactics. This operation exemplifies how modern technologies can be exploited for illegal activities, with ChatGPT being used as a tool to enhance these schemes. Many individuals have suffered as victims of these scams, highlighting the severity of the issue.

Governance

Apple is getting this wrong

3 Aug 202620 Safar 1448 AH

Governance

3 Aug 202620 Safar 1448 AH

Apple is getting this wrong

OpenAI issued a statement responding to Apple's lawsuit, labeling it as baseless. The documents provided by OpenAI include messages that clarify the events and refute claims regarding its employees. This move comes at a critical time as OpenAI aims to protect its reputation and assert its integrity in the face of legal challenges.

Models

The Inference Engineering Masterclass — Philip Kiely & Ali Taha, Baseten

3 Aug 202620 Safar 1448 AH

Models

3 Aug 202620 Safar 1448 AH

The Inference Engineering Masterclass — Philip Kiely & Ali Taha, Baseten

Baseten has raised an impressive $13 billion in a funding round, positioning itself among the leading AI infrastructure decacorns. The company is a key beneficiary of the shift towards inference engineering, which has emerged as a critical discipline in AI. This field focuses on transforming training weights into fast, reliable, and cost-effective products. These developments highlight the significance of inference engineering in enhancing model performance, paving the way for new innovations. As demand for intelligent solutions grows, Baseten is set to play a pivotal role in shaping the future of AI.

Models

Design Arena creators raise $7.9 million to bring taste to AI models

3 Aug 202620 Safar 1448 AH

Models

3 Aug 202620 Safar 1448 AH

Design Arena creators raise $7.9 million to bring taste to AI models

Design Arena is utilized by 5.3 million users globally, providing critical human evaluations to frontier labs. This platform enhances research quality by delivering accurate and reliable assessments, which boosts labs' ability to achieve better outcomes. It is deemed essential across various fields, making it a preferred choice for many institutions. These figures highlight Design Arena's significance in fostering innovation and scientific research, opening new avenues for strategic decision-making in the global market.