Meta Superintelligence Labs released Muse Glimmer, a 30-billion-parameter open-weight model under the Apache 2.0 license, designed to run AI agents locally on a single consumer GPU (24-32GB VRAM). Quantized to under 20GB, it was tested on MacBook M4/M5 Max and NVIDIA RTX 5090. CEO Mark Zuckerberg confirmed the frontier-class Muse Spark 1.2 weights will be released next, marking Meta's return to open weights after the Llama 4 flop.
Nvidia released Nemotron 3.5 Lightning, its first open-source AI model, a 30-billion-parameter model built for single-GPU agentic AI workloads. The company is also developing Nemotron 4, a family of models with the largest version expected to exceed 1 trillion parameters, aimed at challenging top open-source models globally. Nvidia could release the most advanced version as early as late 2026.
xAI released Grok 4.6 on August 12, matching GPT-5.6 Sol for third place on the Artificial Analysis Intelligence Index and overtaking Kimi K3. The model reused the 1.5-trillion-parameter base from Grok 4.5 but delivered a flagship upgrade through post-training alone. On AA-Briefcase workloads, Grok 4.6 completed tasks in roughly 53 turns and 0.5 billion input tokens, versus approximately 103 turns and 2 billion tokens for Claude Opus 5.
Mistral AI announced plans to scale European inference to 1 gigawatt by 2030, up from under 200 MW today across three sites. The company launched Mistral Regional Endpoints (US or Europe), secured five-year contracts with Amadeus, ASML, CMA CGM, and Capgemini, and will host China's GLM-5.2 (by Z.ai) as the first third-party open-weight model on its platform.
A supply chain attack on LiteLLM, a popular open-source LLM gateway, exposed over 2,500 organizations and 434,000 CI/CD pipelines. The breach was enabled by a compromise of Aqua Security's Trivy open-source scanner, which was used to distribute malicious LiteLLM PyPI packages that stole cloud credentials, SSH keys, and Kubernetes tokens.
Alibaba's Qwen3.8-Max scored 56 on the Artificial Analysis Intelligence Index, a 10-point jump over Qwen3.7 Max (46), claiming to beat Claude Opus 4.8 and split with GPT-5.6 Sol on key benchmarks. The open-weight model is priced at $2 per million tokens, though Alibaba will require large commercial users to share revenue.
Mark Zuckerberg published a 6,500-word essay arguing against AI regulation and for open-weight models, calling for "personal superintelligence" over institutional AI. He urged the US to reduce policy barriers for open-source AI to compete with Chinese rivals, while also calling for global regulation of physical AI components to limit harmful material production.