AI News Brief

1. Z.ai's Open-Weight GLM-5.2 Disrupts the AI Race

Zhipu AI (Z.ai) released GLM-5.2, an open-weight model with MIT licensing, a 1-million-token context window, and pricing at $1.40/$4.40 per million tokens - roughly one-sixth the cost of GPT-5.5. It scores 62.1% on SWE-bench Pro (beating GPT-5.5's 58.6%), 81.0 on Terminal-Bench 2.1 (within 4 points of Claude Opus 4.8's 85.0), and ranks #2 on Code Arena behind Claude Fable 5. It runs on Huawei silicon and climbed above multiple Anthropic models on OpenRouter usage rankings.

2. Z.ai Launches Free ZCode Coding Tool to Challenge Cursor and Copilot

Z.ai debuted ZCode, a free AI coding assistant powered by GLM-5.2, positioning it as a direct competitor to Cursor, Claude Code, and GitHub Copilot. The tool combines AI agents with existing developer workflows for planning, coding, reviewing, and deploying. Z.ai called it the "official development environment" for GLM-5.2, undercutting US rivals on pricing.

3. US Government Nears Voluntary AI Model Release Standards

The US Commerce Department is in advanced talks with AI labs to establish voluntary standards for new model releases, with an announcement possible as soon as next week. The framework would set benchmarks, timelines, and access rules for advanced models. This follows a June executive order directing agencies to test frontier models before public release. The Commerce Department also lifted export controls on Anthropic's Fable and Mythos models this week, less than three weeks after suspending them over national security concerns.

4. OpenAI Proposes 5% Equity Stake to Trump Administration

OpenAI has proposed giving the US government a 5% equity stake (valued at approximately $42.6 billion) to ease regulatory pressure, and is urging other AI companies to do the same. The proposal comes as OpenAI agreed to a restricted initial release of GPT-5.6 to government-approved partners while officials evaluate security implications. Trump previously called government ownership stakes in AI firms "a beautiful thing."

5. Meta's "Watermelon" Model Matches GPT-5.5 on Benchmarks

Meta AI head Alexandr Wang told employees that the company's next model, codenamed "Watermelon," has already caught up to GPT-5.5 on closely watched benchmarks. Watermelon is the successor to "Avocado" (the internal name for Muse Spark, released in April) and uses an order of magnitude more compute. Separately, Meta is building an AI cloud business to sell excess compute capacity, with shares jumping 9% on the news.

6. Anthropic Explores Custom AI Chip With Samsung

Anthropic is in early talks with Samsung Electronics to manufacture a custom AI chip, joining OpenAI, Google, Amazon, Microsoft, and Meta in the push to reduce dependence on Nvidia. Anthropic has hired chip engineers including Clive Chan, formerly of Tesla and OpenAI's custom chip teams. Samsung offers a 4nm node (already used in Nvidia's LPU 30) and a newer 2nm SF2P process entering production later this year. This follows OpenAI's Broadcom-built "Jalapeno" chip announcement.

7. Nvidia Launches Revenue-Sharing Model for AI Startups

Nvidia introduced a new business model granting token credits to developers in exchange for a share of future sales, aiming to widen access to its AI hardware for startups lacking capital. The company will connect AI data center operators with cloud service providers and earn both standard product revenue and a share of cloud revenue on supported capacity. The move targets researchers and nascent companies that historically couldn't access large-scale AI compute.