Anthropic officially announced that starting July 20, Claude Fable 5 will be included in all Max and Team Premium plans, with a quota equal to 50% of the subscription allowance. Pro and Team Standard users can still access it via credits, and will receive a one-time $100 credit bonus. Multiple bloggers believe this move is driven by competitive pressure from GPT 5.6 and Kimi K3. Previously, Fable’s subscription quota support was briefly removed in Claude Code, and Anthropic later confirmed it was a bug that has since been fixed.<\/p>\n
Sources:<\/p>\n
The K3 model released by Moonshot AI continues to grow in popularity. Data from DataCurve shows K3 ranks third on DeepSWE, becoming the first open-weight model to reach cutting-edge performance levels. Former AMD executive Patrick Moorhead claimed K3 was distilled from Fable, but multiple bloggers refuted this, noting Fable was only lifted from restrictions on July 1, making data generation and training completed within 15 days technically impossible. K3’s internal testing speed reaches 40-60 token\/s, but has slowed due to a surge in user traffic. Its promotional video was produced by Dubai-based Studio Archive.<\/p>\n
Sources:<\/p>\n
According to The New York Times, Anthropic has proactively approached Meta to lease data center computing power on a monthly basis over the next two years, with a total agreement value of up to $10 billion. Talks are still in the early stages. Due to Meta’s own model capability gaps, its large-scale data centers have become available for external sale. Mark Zuckerberg revealed that companies inquire about purchasing computing power almost every week. Meta is recruiting executive Dave Brown from AWS to advance its cloud business. Anthropic already sources computing power from AWS, Google, Microsoft, SpaceX and other providers.<\/p>\n
Sources:<\/p>\n
Multiple bloggers have discussed AI-driven development models. Blogger Baoyu shared his workflow: using Claude Design to generate high-fidelity interactive prototypes instead of traditional product documents, with AI implementing the final code, achieving over 90% accuracy to the original design. He recommends using prompts to guide agents to automatically generate test code, and leveraging the Desktop version of Computer Use for automated testing. He noted his daily development role is now closer to a product manager and QA: defining functions and verifying results, no longer reviewing every line of code. Developer steipete’s practical comparison found OpenAI Terra high performs significantly better than Sol low in code review scenarios.<\/p>\n
Sources:<\/p>\n
Elon Musk revealed that xAI’s 2-trillion parameter model will complete initial training next week. It outperforms the current 1.5T model in all aspects, is expected to surpass Kimi K3, while maintaining speed and token efficiency close to Grok 4.5. Musk has repeatedly promoted Grok Build and Grok 4.5, noting Grok Build improves almost every day. Grok 4.5 entered the FrontierCode leaderboard, and was rated the most cost-effective coding model by Cognition.<\/p>\n
Sources:<\/p>\n
Moonshot AI CEO Yang Zhilin shared the K2.5 technical roadmap at GTC 2026: MuonClip replaces the Adam optimizer, delivering training results close to doubling the training data on the same dataset; Kimi Linear (KDA linear attention) is the first to outperform full attention across short-context, long-input and long-output tasks; Agent Swarm supports up to 300 agents working in parallel; Attention Residue replaces residual connections with an attention mechanism, bringing approximately 24% improvement in token efficiency. All three components have been open-sourced. He also shared an unexpected finding from early multimodal fusion training: training only on vision tasks can also improve text reasoning ability.<\/p>\n
Sources:<\/p>\n
Codex lead thsottiaux confirmed GPT-5.6 may accidentally delete user files during programming tasks: when Full Access is enabled, and sandbox protection and automatic review are disabled, the model attempts to rewrite the $HOME environment variable during cleanup, and an operational error leads to directory deletion. OpenAI is updating prompt messages, guiding users to safe permission modes, and adding additional security checks.<\/p>\n
Sources:<\/p>\n
Statistics: Number of scanned timelines = 360, Number of matched bloggers = 37, Total matched tweets = 244, Weighted tweet score = 198.4, Number of original tweets = 117, Number of retweeted tweets = 42, Crawl attempts = 2, Boundary coverage status = tail_confidently_crossed_target_boundary<\/p>\n","protected":false},"excerpt":{"rendered":"
Claude Fable 5 adjusts its pricing strategy, the open-source Kimi K3 model sparks global technical discussions, and AI-assisted development showcases great automation potential.<\/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-1473","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\/1473","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=1473"}],"version-history":[{"count":1,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/posts\/1473\/revisions"}],"predecessor-version":[{"id":1474,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/posts\/1473\/revisions\/1474"}],"wp:attachment":[{"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/media?parent=1473"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/categories?post=1473"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/tags?post=1473"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}