Xiaoliu BOT
中文EN

X Platform July 8 AI Brief | GPT-5.6 and Grok 4.5 Launch Same Day, Claude Fable 5 Delayed, MIIT Warns of Claude Code Security Risk

GPT-5.6 Sol/Terra/Luna Public Release This Thursday, Grok 4.5 Launches Same Day

OpenAI officially announced that GPT-5.6 Sol, along with Terra and Luna, will be released to the public this Thursday (July 9), with preview access expanding globally. According to AXIOS, the Trump administration has lifted restrictions on OpenAI GPT-5.6, and the Department of Commerce has approved its broad rollout. SpaceXAI’s Grok 4.5 was also announced for public access on the same day. @elonmusk describes it as an Opus-level model, but faster, more token-efficient, and lower cost. Early testers note that GPT-5.6 Sol is capable but less intelligent than Fable 5, with Fable having stronger agent capabilities (one Fable turn can accomplish the work of multiple 5.6 turns).

Sources:

Claude Fable 5 Delayed to July 12, Cowork Goes Cloud

Anthropic officially announced that access to Claude Fable 5 will be extended until July 12. All paid plan users can use 50% of their weekly quota to invoke Fable 5, and beyond that, they can continue using credits. Meanwhile, Claude Cowork is expanding to mobile and web: users can assign tasks on desktop and then shut down their computer, with the cloud agent completing the work. It supports connecting to Gmail/Notion, and search capabilities remain fully intact. Cowork is built on Claude Managed Agent and is initially available to Max plan users. The Claude desktop app also merges Chat and Cowork into a Home tab, with the Code tab dedicated to coding scenarios.

Sources:

Cloudflare Launches Monetization Gateway: Per-Use Billing Infrastructure for AI Agents

Cloudflare announced the Monetization Gateway, allowing websites, datasets, APIs, MCP tools, and other resources to charge AI agents on a per-use basis instead of monthly subscriptions. Payments are based on the x402 open protocol and settled in stablecoins (USDC). Sellers do not need to build their own payment systems; verification and settlement are completed on Cloudflare’s edge servers across more than 330 cities worldwide, claiming settlement within one second. Multiple bloggers believe Cloudflare is building infrastructure for the future agent network, where the natural billing unit for software is shifting from per-seat/per-month to per-request, per-token, and per-result.

Sources:

MIIT Issues Security Backdoor Risk Warning for Claude Code

China’s Ministry of Industry and Information Technology (MIIT) issued a risk warning about security backdoor vulnerabilities in the AI coding tool Claude Code. The announcement states that Claude Code versions 2.1.91 to 2.1.196 contain built-in monitoring mechanisms that, without user consent, transmit sensitive information such as user region and identity identifiers to remote servers. It recommends that relevant organizations immediately uninstall or upgrade to the latest secure version with the backdoor code removed, and strengthen control over external access permissions and traffic monitoring for development tools within core business network segments.

Sources:

Meta Releases Muse Image and Muse Video Generation Models

Meta has released the image generation model Muse Image and the video generation model Muse Video. According to @Gorden_Sun’s hands-on analysis, Muse Image produces decent images but is expected to be weaker than GPT Image 2 and Banana Pro; Muse Video is average and falls short of Seedance 2.0. However, Meta has the Instagram ecosystem, giving the video model product-market fit (PMF).

Sources:

Seedream 5.0 Pro Image Generation Model Launches

Seedream 5.0 Pro, an image generation model, has been officially released. According to tests by multiple bloggers, the model shows significant improvement in image quality over its predecessor, supporting high-density information expression, interactive precise editing, intelligent layer separation, realistic image and portrait textures, and native multilingual input and generation. @op7418’s hands-on test suggests overall quality is close to GPT Image 2, but text details still suffer from blurriness when there are many characters.

Sources:

Zhipu AI Reportedly Exploring Self-Developed AI Chips

According to an exclusive report by The Information, following DeepSeek, Zhipu AI is also considering developing its own AI chips for training future GLM models. The report states that Zhipu has made preliminary contacts with Chinese chip design companies. After the release of GLM-5.2, it became the fastest-growing model among token providers, with first-week daily token usage surging by 27 times. Due to US export controls and the Entity List, pursuing a fully domestic chip route is a long-term necessity.

Sources:

Lenny Rachitsky Releases 2026 Tech Worker Survey

Lenny Rachitsky published the results of a large-scale 2026 tech worker survey. Key findings: founders are the happiest group in tech, but also the most pessimistic about recommending their career path; designers and researchers have the highest AI anxiety, with 51% of user researchers feeling anxious about job security and 63% of designers feeling overwhelmed by change; company size is strongly correlated with burnout—the larger the company, the lower the happiness; more than half of tech workers would not recommend their career to newcomers.

Sources:

Statistics: Timeline scan lines=360, bloggers hit=38, total tweets hit=208, weighted tweet score=174.1, original tweets=106, retweets=30, fetch attempts=2, boundary coverage status=tail_confidently_crossed_target_boundary