01
GLM-5.3 Launches: Z.ai's Open-Weight Model Beats Anthropic on CyberGym, Weights Coming in Two Weeks
Z.ai (Zhipu) released GLM-5.3 on August 14, built on the same 743B-parameter base model as GLM-5.2 with all gains from scaled-up post-training. It scores 84.5% on CyberGym, surpassing Anthropic's Mythos 5 and OpenAI's GPT-5.6 Sol, and improves coding 50% over GLM-5.2 with 66.9% on DeepSWE and 28.3 on Terminal-Bench 3.0 (up from 4.6). The model found 1,097 critical vulnerabilities in Linux, WebKit, and FreeBSD. Open weights arrive on Hugging Face in roughly two weeks after a safety assessment, with API access priced at about a tenth of US frontier per-token rates.
02
DeepSeek Ships V4 Pro Model and Open-Source Harness Rival to Claude Code
DeepSeek formally released its official V4 Pro model (V4-Pro-0813) on August 14, enhancing AI agent and software engineering capabilities, available via website, mobile app, and API. Alongside it, DeepSeek launched Harness, an open-source coding agent rivaling Claude Code. The company raised prices up to 4x on V4 Flash and Pro models, though blended V4 Pro pricing remains below GLM-5.2. The cheaper V4 Flash had unexpectedly outperformed the April V4 Pro preview in several independent tests.
03
Alibaba's Qwen Crosses 3 Billion Downloads, Surpassing Meta and Google
Alibaba's open-weight Qwen model family has accumulated over 3 billion global downloads in the past six months, eclipsing Meta (227 million) and Google (418 million) to become the world's most-downloaded AI model family per Hugging Face's state of open models report. Alibaba also released Qwen3.8-27B on August 14, an open-weight 27B-parameter model with a 262k token context window, scoring 90.3% on LiveCodeBench v6 and 89.2% on GPQA Diamond.
04
Google Launches Gemini 3.7 Flash at Half Price, Claims Edge on Business Workflows
Google released Gemini 3.7 Flash with a 50% introductory price cut through December 31, 2026, bringing token costs to $0.75 input / $3.75 output per million tokens. The model scores 30.4% on AutomationBench (up from 17.0% for Gemini 3.6 Flash), which Google says outperforms Claude Sonnet 5 (10.6%) and GPT-5.6 Terra on business workflow automation. It also reaches the Intelligence vs. Time per Task Pareto frontier with a 4-point improvement on the Intelligence Index over its predecessor.
05
White House Eyes Adding Open Models to AI Framework as FTC Moves on AI Accuracy
The White House is considering adding open models to an updated voluntary AI framework, sources tell WIRED, as it continues to grapple with regulating a technology Trump has refused to formally regulate for fear of helping China. Separately, the FTC issued a proposed policy statement on AI accuracy and output steering under Executive Order 14365, while the Cato Institute criticized the administration for not releasing its frontier model testing framework publicly.
06
Nvidia Enlists Wall Street for $500 Billion AI Infrastructure Financing Push
Nvidia signed memorandums of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to mobilize over $500 billion in third-party capital for AI data centers, chip factories, and power stations. CEO Jensen Huang declared Nvidia's chips are now an "investable asset class," marking a shift from project-by-project data center builds to infrastructure-scale financing. Bank of America simultaneously elevated its AI chip forecast and selected AMD and Nvidia as top semiconductor picks, upgrading its 2030 server CPU forecast to $210 billion.
07
Zuckerberg Publishes 6,500-Word Essay Advocating Open AI and Global Regulation
Meta CEO Mark Zuckerberg published a 6,500-word essay detailing his vision for AI, warning against centralizing AI power and calling for global regulation focused on limiting the physical production of harmful materials rather than restricting software or open weights. Zuckerberg argued that restricting open weights would concentrate power in a small number of institutions, positioning Meta's open-weight strategy as a counterweight to closed-model dominance from OpenAI and Anthropic.