[AINews] High Taste Testers Ep. 1

Latent Space is testing a new series called High Taste Testers, which blends informal food conversations in San Francisco with the latest AI topics, such as the recent Dev Day and cutting‑edge assistant models. In the edition covering the week of 15‑16 October 2026, the AI News bulletin reviews the most significant open‑source model releases in the United States. Meta unveiled Muse Spark Open, an expert‑mixing (MoE) model with 816 billion total parameters and 42 billion active, 62 layers, and multimodal capability (text, image, video, and audio). According to Artificial Analysis, it scores 39 on its Intelligence Index, surpassing NVIDIA’s Nemotron 3.5 Super VL. The estimated cost per indexing task is $0.84, lower than several competitors, though its agent‑evaluation performance lags behind other open models. Later that day, Reflection Beam 1.0, another MoE with 501 billion total and 23 billion active parameters, launched under an Apache 2.0 license and earned a score of 30 on the same index, ranking it as the second‑highest‑scoring open model in the U.S. Both models receive immediate support in tools such as mlx‑lm, vLLM, and SemiAnalysis, and are available on platforms like Ollama with zero data‑retention conditions.

These developments are significant because they broaden access to large‑scale models at more competitive prices, potentially accelerating the adoption of generative AI in business and research applications and fostering greater diversity in the open‑source ecosystem.

Read the original article on Latent Space

This summary is an informational synthesis produced by dataqbs.com. All rights to the original content belong to its author and the cited media outlet. We act solely as curators of technology news and claim no authorship.