OpenAI Unveils First Self-Developed AI Chip, Jalapeño
OpenAI has officially released its first self-developed AI chip, Jalapeño, manufactured by Broadcom. It is specifically designed for LLM workloads powering ChatGPT, Codex, API, and future Agent products. OpenAI states that developing its own chip will expand its full-stack platform from products to models and infrastructure. Following the announcement, Cerebras’ stock price plummeted. Bloggers note that chips are the infrastructure of the AI economy, and in-house development helps reduce reliance on third-party suppliers and increases control over computing power.
Sources:
- @OpenAI: https://x.com/OpenAI/status/2069770172802773292
- @Gorden_Sun: https://x.com/Gorden_Sun/status/2069779546317164654
Anthropic Launches Claude Tag: An AI Colleague in Slack
Anthropic has released Claude Tag, enabling Claude to reside as a permanent colleague within Slack channels. Team members can assign tasks by @mentioning Claude in a channel; Claude will then complete the tasks step-by-step and deliver results within the thread, powered by the underlying Opus 4.8 model. Karpathy praised this as the third major evolution in LLM UI/UX—transitioning from websites and local applications to an organizational-level, persistent collaborative entity. Multiple bloggers analyzed the product’s significance from various angles: the design philosophy of using channels as context boundaries differs from traditional role-based or team-context approaches; Claude Tag supports an ambient mode for proactively pushing information and following up on unanswered threads.
Sources:
- @claudeai: https://x.com/claudeai/status/2069468693017268244
- @dotey: https://x.com/dotey/status/2069477417278730536
- @karpathy: https://x.com/karpathy/status/2069547676849557725
- @oran_ge: https://x.com/oran_ge/status/2069534869047447920
- @Gorden_Sun: https://x.com/Gorden_Sun/status/2069631540611691003
ByteDance’s Multi-Product Intensive Updates: Doubao Pro, Seedance 2.5, SeedMusic Preview
ByteDance’s AI products have received multi-threaded updates. Doubao launched a Pro version, initiating a paid subscription (starting at 68 CNY per month for continuous billing, up to 500 CNY/month). It is powered by the latest Seed 2.1 Pro model and supports autonomous execution capabilities like operating local computers, browsers, Skills, and scheduled tasks. The video model Seedance 2.5 was released, supporting single-generation of 30-second short videos and native 4K resolution. The music model SeedMusic 1.0 Preview went live, with bloggers reporting good results but noting that vocals still retain an electronic tone. The Seed 2.1 Pro model has been launched on the Cola platform, with multiple bloggers rating it as the current strongest multimodal model.
Sources:
- @xiaohu: https://x.com/xiaohu/status/2069593382247444962
- @xiaohu: https://x.com/xiaohu/status/2069781928056902121
- @xiaohu: https://x.com/xiaohu/status/2069655846896935269
- @vista8: https://x.com/vista8/status/2069580863457203731
- @oran_ge: https://x.com/oran_ge/status/2069721885508870274
- @Gorden_Sun: https://x.com/Gorden_Sun/status/2069673054314389725
- @Chengzilhy: https://x.com/Chengzilhy/status/2069765583189803478
Starlink Accelerates Deployment in Aviation and Emergency Communications
Starlink saw multiple advancements in aviation and emergency communications on the same day. American Airlines confirmed it will begin deploying Starlink on over 500 narrow-body aircraft starting early 2026; Southwest Airlines’ first Starlink-equipped flight has already taken off. Starlink will also provide emergency communication support for all 78 municipalities in Puerto Rico, as announced by the Governor of Puerto Rico. Additionally, SpaceX’s “Starfall” project will offer a microgravity environment for scientific research and space manufacturing, and SpaceX’s AI satellite constellation has been officially named “Starmind.”
Sources:
- @elonmusk: https://x.com/elonmusk/status/2069535601892987275
- @elonmusk: https://x.com/elonmusk/status/2069807896767107532
- @elonmusk: https://x.com/elonmusk/status/2069484512984371663
- @elonmusk: https://x.com/elonmusk/status/2069493276525449553
- @Starlink: https://x.com/Starlink/status/2069456317635498279
OpenAI Increases Open-Source Ecosystem Investment & Opens DevDay 2026 Applications
OpenAI Developers has released 30+ models and features for its API over the past six months. Codex for Open Source has provided six months of ChatGPT Pro access to over 3,500 maintainers. OpenAI has joined the Rust Foundation as a Platinum Member, providing funding to support the Rust language ecosystem, and launched the Patch the Planet project (AI-assisted open-source security research). DevDay 2026 applications are now open; the event will be held in San Francisco on September 29th and expanded to cities including Bangalore, Tokyo, Seoul, Paris, Berlin, London, São Paulo, and Mexico City.
Sources:
- @OpenAIDevs: https://x.com/OpenAIDevs/status/2069499656305090671
- @OpenAIDevs: https://x.com/OpenAIDevs/status/2069457053035438427
- @OpenAIDevs: https://x.com/OpenAIDevs/status/2069457055227940891
- @OpenAIDevs: https://x.com/OpenAIDevs/status/2069484303281779090
- @OpenAI: https://x.com/OpenAI/status/2069483224158646739
Krea 2 Open-Source Image Model Weights / Baidu Unlimited OCR / Higgsfield Growth Story
Krea released open weights for Krea 2, including Raw and Turbo versions. The former is suitable for fine-tuning, while the latter is a fast, distilled version. The license permits commercial use by small companies and individuals. Baidu released an Unlimited OCR model based on R-SWA (Reference Sliding Window Attention) technology, capable of parsing dozens of document pages in a single forward inference, achieving 93% accuracy on OmniDocBench v1.5. The founder of Higgsfield shared the growth journey: reaching over 24 million users and $400 million ARR within 13 months, making it the fastest-growing video AI startup.
Sources:
- @krea_ai: https://x.com/krea_ai/status/2069435590995812396
- @MaxForAI: https://x.com/MaxForAI/status/2069466097624199576
- @xiaohu: https://x.com/xiaohu/status/2069630326641758518
- @MaxForAI: https://x.com/MaxForAI/status/2069481623721329032
Google Workspace CLI Author Fired for Creating the Project
Google veteran Justin Poehnelt, a seven-year employee, was fired by Google for creating the Google Workspace CLI (written in Rust with a built-in MCP server). The project topped Hacker News after its release in March, garnering over 26,000 GitHub stars and thousands of actual users. During the Google Cloud Next conference, Google officially announced the launch of Workspace CLI; two days later, the author was terminated. Multiple bloggers discussed the controversy: some view it as a classic case of Google’s bureaucracy stifling innovation, while others point out that releasing an open-source project using the company’s brand without approval could lead to termination at any large company. The project continues to be maintained normally under the Google GitHub organization.
Sources:
- @dotey: https://x.com/dotey/status/2069537208378302874
- @steipete: https://x.com/steipete/status/2069594195522941059
- @JPoehnelt: https://x.com/JPoehnelt/status/2069482265953087602
Stats: Timeline Scans=360 Bloggers Mentioned=41 Total Tweets Matched=212 Weighted Tweet Score=165.05 Original Tweets=91 Retweets=46 Crawl Attempts=2 Boundary Coverage Status=tail_confidently_crossed_target_boundary