Latest results from DeepSWE Bench show GPT-5.5 leading with 70% Pass@1, far ahead of Claude Opus 4.8’s 58%, securing first place on the benchmark. DeepSWE Bench is a long-cycle, high-difficulty coding benchmark, sparking discussions among multiple bloggers. @MaxForAI directly said “Who still uses Claude?”, @steipete retweeted the news of GPT-5.5 topping the chart. @thsottiaux (Codex lead) also referenced the discussion in a tweet, saying “5.5 xhigh at 70%, $6.6”. This round of benchmarks further solidifies GPT-5.5’s leading position in the coding agent field.<\/p>\n
Sources:<\/p>\n
According to Axios, NVIDIA and Microsoft are expected to unveil the first Windows PCs powered by NVIDIA chips during Computex next week, including product lines such as Surface and Dell. This is not just “equipped with NVIDIA graphics cards”\u2014NVIDIA is entering the Windows PC core processor market, with plans to release Arm-based N1\/N1X laptop processors. Microsoft, NVIDIA, and Arm have simultaneously teased “A new era of PC.” If officially realized, the Windows camp will see a new competitive landscape involving Intel, AMD, Qualcomm, and NVIDIA. This marks a key step in NVIDIA’s expansion from data center AI chips to the personal computing market.<\/p>\n
Sources:<\/p>\n
Google DeepMind recently released a major podcast featuring four core Gemini members: Jeff Dean, Noam Shazeer (co-author of the Transformer paper), Koray Kavukcuoglu (DeepMind CTO), and Oriol Vinyals, discussing Gemini’s past and future. Key takeaways: First, the most important direction in the coming year is self-learning\u2014letting the model participate in improving Gemini itself, expected to begin around 2027. Second, long-running agents are a clear frontier but require memory systems, better hardware, and toolchain adaptation\u2014Jeff Dean noted that many current tools are designed for human latency, causing agents to stall during high-frequency calls; Noam said “in a 30-day task, maybe 29.5 days are spent waiting.” Third, the ability to consistently compress Pro-level capabilities into Flash is their most surprising progress; the next-generation Flash sometimes even surpasses the previous Pro. Fourth, the ultimate direction of multimodality is a world model\u2014understanding all modalities of the physical world, not just text and images. @MaxForAI provided a detailed Chinese summary of this podcast.<\/p>\n
Sources:<\/p>\n
Codex lead @thsottiaux announced that Codex has reached 5 million users and reset usage limits for all paid ChatGPT subscribers. The previous 10x usage benefit for Pro users has also expired. Many bloggers celebrated the reset; @MaxForAI exclaimed “Let’s go!” and @dotey also retweeted the news. @thsottiaux said “Let the tokens do incredible things today.” Codex GPT-5.5 is currently considered by many users as the strongest coding model, and the 5 million user milestone confirms its rapid growth in the coding agent market.<\/p>\n
Sources:<\/p>\n
A demo video of GPT-Realtime 2.0 has gone viral on social media: users control a computer in real time using only voice, without any manual operation. @FarzaTV released the full demo, and @xiaohu retweeted it, commenting “This is the real Siri.” Meanwhile, OpenAI Developers announced the four finalist projects for Voice Hack Night: Agentic OS for a Phone (voice-first mobile OS), Wagner (multi-agent virtual meeting room), and Curo (one-on-one AI tutoring), showcasing the application capabilities of real-time voice agents in practical scenarios. Voice as a human-computer interaction interface is moving from concept to usable product.<\/p>\n
Sources:<\/p>\n
Weijie Su, a full professor at the Wharton School of the University of Pennsylvania, announced he is joining OpenAI during his sabbatical. He holds a BS in CS+Math from Penn and a PhD from Stanford, and had already earned tenure at Wharton. @MaxForAI commented after citing the news: “Ten years ago, the smartest people stayed in academia to publish papers; now, the smartest people find industry problems more interesting, with more resources and faster deployment.” @dotey also retweeted the news. This continues the trend of the AI industry attracting top talent from academia.<\/p>\n
Sources:<\/p>\n
The Tianya community, which has been offline for three years, announced it will officially resume access on June 1, 2026. According to a joint announcement, “Chengdu Tianyake Network Technology Co., Ltd.,” founded in 2024, has invested over one million yuan in preserving Tianya’s data and launched the “New Tianya Genesis Member Product Service Package,” limited to 9,999 units at 1,999 yuan each. Proceeds will fund data preservation and access restoration. Benefits include digital badges, premium membership gift boxes, and 10-year free reading access to the premium post section. The Tianya community has been inaccessible since April 2023 due to unpaid telecom IDC fees. This relaunch has sparked nostalgia among many Chinese bloggers. @imwsl90 exclaimed, “What year is this?”<\/p>\n
Sources:<\/p>\n
@dotey proposed a systematic view on Agent OS: the universal agent is the future operating system, and current apps will face three outcomes\u2014die out (if the agent can do it itself), become CLI\/MCP (for agent invocation), or become Agent GUI plugins (temporarily invoked when manual operation is needed). He further pointed out that the SaaS industry will see a trend: everyone must launch CLI + Skill to let agents learn to call them, or they will be eliminated. @xiaohu observed that “everyone in AI is a player”\u2014Vibe Coding users frequently switch between Claude Code and Codex, reflecting the intensity of tool competition.<\/p>\n
Sources:<\/p>\n
Statistics: Number of scan timeline lines=360, Number of bloggers hit=31, Total tweets hit=147, Weighted tweet score=119.25, Original tweets=72, Retweets=27, Fetch attempts=2, Boundary coverage status=tail_confidently_crossed_target_boundary<\/p>\n","protected":false},"excerpt":{"rendered":"
GPT-5.5 significantly leads in coding benchmarks, NVIDIA announces entry into PC processor market, and universal agent is discussed as the core form of future operating systems.<\/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-1313","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\/1313","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=1313"}],"version-history":[{"count":0,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/posts\/1313\/revisions"}],"wp:attachment":[{"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/media?parent=1313"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/categories?post=1313"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/tags?post=1313"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}