Alibaba has introduced Qwen3.8-Max, its largest AI model yet, featuring 2.4 trillion parameters and a mixture-of-experts architecture to optimize cost and performance. DeepSeek's V4-Flash model, with 284 billion parameters, offers lower inference pricing, challenging competitors in the market. Both companies are promoting open-weight models, allowing developers to deploy AI on their own infrastructure. This move underscores a trend towards more cost-effective and flexible AI solutions in China, as companies like Alibaba and DeepSeek compete on both performance and price.
Read originalStanford's Evo 2 AI model has made a significant leap in synthetic biology by generating phages that effectively target E. coli. This breakthrough demonstrates the potential of AI to design entire viral genomes, moving beyond simple DNA edits. The model produced thousands of candidate genomes, with 16 showing strong E. coli-killing activity in lab tests. By releasing Evo 2 as open-source software, Stanford is inviting further exploration and innovation in genome design, potentially paving the way for new treatments against resistant bacteria like MRSA.
Alibaba is testing a new business model for its upcoming Qwen open-weight AI model, introducing revenue-sharing terms for commercial users. This move targets larger companies that profit from offering the model as a service, requiring them to enter a commercial agreement with Alibaba. The approach mirrors the licensing model used by Moonshot for its Kimi K3 model, which includes revenue-sharing for companies exceeding certain revenue thresholds. This shift signifies Alibaba's strategy to monetize its open-weight models while maintaining their open-source nature, potentially setting a precedent for other AI developers.
The b10311 release of llama.cpp tackles inefficiencies in text-to-speech (TTS) generation by refining how text streams are processed. Previously, the system would redundantly handle utterances, causing them to be read twice before completion. This update aligns the streaming overlay with the non-streaming prefill, effectively eliminating the duplication. Developers working with TTS systems will find this change streamlines the generation process and boosts efficiency. The update is accessible on macOS, Linux, and Windows, ensuring that a broad range of users can benefit from these improvements.
The b10313 release of llama.cpp introduces an LRU scheduler, significantly enhancing task management efficiency. This update includes improvements in handling coalescing, optimizing the waiting queue, and fixes for stream cases to ensure smoother operations. The release also expands platform-specific builds, such as Vulkan and ROCm 7.2 support on Ubuntu, and CUDA 12 and 13 on Windows. While there are no new model architectures, these updates demonstrate a commitment to refining performance and compatibility across various systems.