Competition in the large AI model space reached a fever pitch yesterday. OpenAI’s product lead, Tibo, announced a series of major adjustments: temporarily removing the 5-hour usage limit for all Plus, Business, and Pro plans; simultaneously launching inference efficiency optimizations to make GPT-5.6 Sol more resource-efficient across scenarios; and pushing quota resets to approximately 500,000 ChatGPT Work and Codex users. CEO Sam Altman publicly solicited interesting projects based on Sol, promising special gifts from the OpenAI archives.<\/p>\n
Anthropic quickly followed suit, expanding access to Claude Fable 5\u2014previously limited to Code subscribers\u2014to all paid plans, extending the offer until July 19, and increasing the Claude Code weekly limit by 50%. Several bloggers commented that the strong pressure from OpenAI’s latest model forced Anthropic to change its strategy.<\/p>\n
xAI was also active. Elon Musk frequently promoted Grok 4.5, claiming it outperforms Fable in some software benchmarks and reaches Opus-level performance in browser usage scenarios. He also revealed that Grok 4.5 is the first collaborative result following SpaceX’s acquisition of Cursor, showcasing improvements to Grok Build’s navigation, keyboard workflow, and background agent mode. Sources:<\/p>\n
Multiple bloggers pointed out that Grok’s Build CLI packages and uploads a user’s entire project codebase, including all tracked files and git history. A developer could potentially leak API keys and sensitive configuration information just by having Grok reply with a single word. Blogger Guizang described this behavior as “way too outrageous,” citing other sources’ findings that it reads .env files, key files, and the complete project structure. Several bloggers advised caution until the issue is fixed.<\/p>\n
Sources:<\/p>\n
Blogger MaxForAI, citing informed sources, reported that Tian Keyu, a former intern who was fired and sued for attacking ByteDance’s large model training tasks, has now started a company in the world model field and raised tens of millions of dollars in funding, with an estimated valuation of around $200 million. According to business registration information, he serves as the legal representative for several related companies, which were established in May 2026 with business scopes covering AI hardware, intelligent robotics, and AI technology R&D. Tian Keyu is a Ph.D. candidate at Peking University and won the NeurIPS 2024 Best Paper Award.<\/p>\n
Sources:<\/p>\n
The OpenAI Codex ecosystem saw multi-directional feature updates yesterday. Inference optimizations have been implemented, significantly boosting efficiency. Video editing capabilities became a highlight, with several bloggers demonstrating Codex’s application in video editing scenarios, including completing edits directly through conversation after installing MCP and Skill. Tom Huang showcased Codex + ChatPPT replicating Manus and Genspark’s PPT generation functionality, with nearly a hundred creative templates and a 95% replication accuracy. OpenAI also officially released nine workflow guides for Codex prompts, covering scenarios like bug fixing, screenshot-to-prototype conversion, and cloud refactoring.<\/p>\n
Sources:<\/p>\n
AI-generated content (AIGC) has taken a substantive step in commercial application. Several bloggers demonstrated Seedream 5.0 Pro’s fine-grained image editing capabilities, enabling precise regional edits with perfect color blending and no overflow. Blogger Xiaoyu Chengzi released a lingerie commercial advertisement entirely created by AI (Seedream 5.0 Pro + Kling 3 Omni), sparking widespread discussion. Additionally, AI-generated photography shared by blogger MANISH1027512 gained nearly 2000 followers on Xiaohongshu\/Douyin in less than 24 hours, demonstrating that AI visual content already possesses natural virality on consumer platforms.<\/p>\n
Sources:<\/p>\n
Several bloggers shared quantified data on AI-assisted development. Barret_China estimated from weekend experiments that with 4 concurrent projects running R&D 7×24, high-quality token consumption costs about $4,000 per month (equivalent to a senior engineer’s salary), while output efficiency is at least 5-10 times that of humans; costs can be further halved as workflows mature. Wesley also shared practical experience using Claude Code to refactor a website, stating that AI compensated for his CSS and design shortcomings, significantly boosting development efficiency. Orange AI shared insights on Codex video editing, announcing plans to open-source related Skills\/Plugins.<\/p>\n
Sources:<\/p>\n
Recent remarks by Microsoft CEO Satya Nadella sparked discussion: when companies call upon AI, they pay API fees while also handing over their internal knowledge. He proposed a “Five C” framework to help enterprises protect and manage organizational experiential knowledge in the AI era. Several bloggers expanded on this theme: Xiangyang Qiaomu pointed out that the truly scarce resources in the future will be the ability to identify worthwhile questions, make judgments with insufficient information, and take responsibility for the consequences; IBM noted as early as 1979 that “computers must never make management decisions,” a principle that holds in the AI era. Another blogger focused on the conflicting interests between business owners and employees regarding skill packaging: owners want to condense employee experience into Skills, while employees fear their value being diluted.<\/p>\n
Sources:<\/p>\n
Stats: Timeline Scans=360 Bloggers Matched=34 Total Tweets Matched=202 Weighted Tweet Score=163.6 Original Tweets=100 Retweeted Tweets=36 Crawl Attempts=2 Boundary Coverage Status=tail_confidently_crossed_target_boundary<\/p>\n","protected":false},"excerpt":{"rendered":"
AI model giants are relaxing restrictions to compete for users, while Grok Build’s privacy issues raise concerns. Meanwhile, AI-generated images and videos are now capable of producing high-quality commercial advertisements.<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[5],"tags":[19],"class_list":["post-1452","post","type-post","status-publish","format-standard","hentry","category-brief","tag-x--ai-"],"_links":{"self":[{"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/posts\/1452","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/comments?post=1452"}],"version-history":[{"count":2,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/posts\/1452\/revisions"}],"predecessor-version":[{"id":1457,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/posts\/1452\/revisions\/1457"}],"wp:attachment":[{"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/media?parent=1452"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/categories?post=1452"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/tags?post=1452"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}