OpenAI secures PORTS-Pike data center capacity deal, with initial deployment expected to reach 4.25 GW
要闻
OpenAI announced a partnership with SB Energy and NVIDIA to lease capacity at the PORTS-Pike data center in Ohio and build and operate AI factories with an initial capacity of 4.25 GW. While securing resources at the site, NVIDIA is also providing guarantees for specified portions of the lease and electricity costs.
OpenAI announced that it has reached an agreement to use capacity at the PORTS-Pike Technology Data Center in Pike County, Ohio, in partnership with SB Energy, NVIDIA, and the U.S. Department of Energy. Jensen Huang said in a post that NVIDIA is working with SB Energy to secure land, power, and data center space at the site, while OpenAI will serve as the tenant and build and operate an AI factory powered by NVIDIA’s full-stack DSX platform. The initial deployment is expected to provide 4.25 gigawatts of capacity, and NVIDIA may expand the arrangement to cover the remaining 3.75 gigawatts. NVIDIA said its support is limited to specified portions of the lease and electricity costs, as well as a residual value commitment. The guarantees will take effect in phases as the data center comes online between 2028 and 2030, with OpenAI responsible for lease payments.
Anthropic announced the launch of /design in Claude Code, bringing Claude Design’s canvas workflow to the CLI and desktop app. Paid subscribers can access the feature after updating Claude Code to the latest version.
Anthropic announced that Claude Code has introduced a new /design skill, currently in research preview. Built on artifacts, the skill brings Claude Design’s canvas workflow to the Claude Code CLI and desktop app. Running /design opens an editable, UI-focused canvas where users can select and refine a design before handing it off to Claude for implementation. Claude reads the codebase, matches the existing UI style, and generates a shareable mock in an Artifact. According to Anthropic, the feature is available to Pro, Max, Team, and Enterprise users, who must update Claude Code to the latest version before using it.
OpenCode raises DeepSeek-V4-Flash allowance to $30
要闻
OpenCode announced the completion of phase one of Operation Cheepseek. Go subscribers can now pay $10 to receive $30 worth of DeepSeek-V4-Flash usage, and phase two is already underway.
OpenCode announced the completion of the first phase of an initiative called Operation Cheepseek. OpenCode Go subscribers can currently pay 10美元 to receive 30美元 in usage credits for DeepSeek-V4-Flash. The second phase is now underway, although the announcement did not disclose any specific details.
Qwen-Audio-3.0 voice model series officially launches on 千问AI平台
模型发布
The Qwen-Audio-3.0 series of voice models is now officially available on 千问AI平台. Developers can access the models via API or by subscribing to a Token Plan.
The Qwen-Audio-3.0 family of speech models has officially launched on the Qwen AI platform. The family includes the Qwen-Audio-3.0-ASR large speech recognition model, the Qwen-Audio-3.0-TTS large speech synthesis model, and the Qwen-Audio-3.0-Realtime real-time voice interaction model. Developers can access the models through API services or subscribe to a Token Plan, which is currently available at a limited-time discount. The models have already been integrated into products including the Qwen App, Qwen Office, and Qoder.
Alibaba launches HappyShrimp 1.0 AI music model in China and overseas
模型发布
Alibaba has unveiled the HappyShrimp 1.0 AI music model, known in Chinese as 快乐虾米. It is now available simultaneously on the eponymous AI music creation platforms in China and overseas, with new users eligible for a large amount of free credits.
Alibaba has released HappyShrimp 1.0, an AI music model whose Chinese name is 快乐虾米, launching it simultaneously in China and overseas starting today. According to Alibaba, the model can accurately understand natural language, allowing users to create a complete song—from lyrics and composition to arrangement and vocals—simply by describing an emotion, a story, or a memory. The model uses an end-to-end, full-song generation approach that creates lyrics, melody, arrangement, and vocals as a unified whole, while supporting precise control through complex, multidimensional instructions covering vocal gender, singing style, key, BPM, and more. Unlike “tag-based input” tools that require users to stack labels such as music genres and instrument names, HappyShrimp enables music creation without professional terminology. On its first day of availability, HappyShrimp announced a strategic partnership with Taihe Music Group. The two companies will explore collaboration in areas including the joint development of the music industry ecosystem, AI music platforms, and co-creation with musicians.
GPT-5.6 Sol goes half-price on OpenRouter and Vercel AI Gateway for one month
开发生态
OpenRouter and Vercel AI Gateway have announced a 50% discount on OpenAI GPT-5.6 Sol, with both promotions running for one month.
OpenRouter and Vercel have each announced a 50% discount on OpenAI GPT-5.6 Sol, with both promotions running through September 18. OpenRouter's 50% discount covers the batch API, flex, and priority (fast) tiers and is applied automatically when using openai/gpt-5.6-sol, but it is limited to the OpenAI provider and does not apply to BYOK requests. Vercel AI Gateway's 50% discount applies to both standard and fast modes.
GitHub experienced a widespread outage affecting eight services, including API, Actions, and Copilot. The company says the incident has been resolved, with a root-cause analysis to be released later.
GitHub's official status page shows that GitHub.com previously experienced a widespread service outage affecting eight services: Git Operations, Webhooks, API Requests, Issues, Pull Requests, Actions, Pages, and Copilot. The incident has now been resolved. During the outage, error rates for web and API traffic were approximately 20%, while error rates for archive downloads and raw repository content downloads were approximately 50%. SAML and OIDC authentication, SCIM, and Team Sync were also affected. After GitHub identified the problematic component and implemented corrective measures, most services gradually recovered. The remaining impact was limited to intermittent Copilot authentication failures in certain applications; using Copilot through GitHub CLI and GitHub App was unaffected. GitHub said a detailed root cause analysis will be published when available.
Cursor launches Origin code hosting platform in early beta for paid users
开发生态
Cursor has launched its Origin code hosting platform, rolling out an early beta to users on all paid plans starting today. The initial release includes repositories, pull requests, and GitHub sync, with Agent-native features coming soon.
Cursor has launched Origin, a code hosting platform now available in early beta to users on all paid subscription plans. Its initial feature set is designed around Agent scale and includes repositories, pull requests, code browsing, and GitHub synchronization, with Agent-native capabilities coming soon. Users can create repositories from the Codebase tab, push local projects through the CLI, or connect GitHub and select repositories to sync. Synced repositories are updated in real time, comments are synchronized bidirectionally, and pushes still go to GitHub. Integrations with Vercel, Depot, and Buildkite are already available, with more coming soon.
千问办公 open-sources MyContext, surpassing 1,000 Stars in one week
开发生态
千问办公 recently open-sourced MyContext, a context infrastructure project. It organizes instant messages, documents, and meeting data into personal context that Agents can use, enabling retrieval and reasoning by large language models and external AI applications.
Qwen Office recently open-sourced MyContext, a context infrastructure project that has received over 1k Star within one week of its release. Its GitHub repository describes it as a local-first desktop application for personal work context. MyContext uses data-source connectors to access authorized data such as instant-messaging conversations, documents, and meeting notes, incrementally collecting and organizing it locally into a continuously updated personal context that can be retrieved and used for reasoning by large models, local Agents, and authorized external AI applications.
Cumora team chat app goes open source, enabling AI Agents to initiate conversations
产品应用
Prominent developer yetone has open-sourced Cumora, a cross-platform team chat app. Its AI Agents have memory and personas and can proactively start conversations and claim tasks. The project is fully open source and supports local deployment, while its cloud service is currently in closed beta.
Developer yetone has open-sourced Cumora on GitHub, a cross-platform team chat application that turns AI Agents into full-fledged members alongside humans, sharing the same member list, direct messages, group chats, kanban boards, and calendars. Each Agent has its own persona, memory, and status, and can proactively initiate conversations, claim tasks, coordinate with others, and send and receive real emails. The project offers two operating modes: Cumora Cloud runs each Agent in a hosted, isolated Pod powered by the OpenAI Responses API; in BYOA mode, users can run a single command to make a local Claude Code or Codex CLI instance the Agent's brain, with their keys never passing through Cumora servers. The project is fully open source and supports local deployment, requiring Postgres and Redis, while its desktop client is available for macOS, Windows, and Linux.
豆包 updates ‘工作任务’ mode with remote PC control from phones
产品应用
豆包 announced an update to its ‘工作任务’ mode that enables remote computer control from a phone. Once authorization is completed, users can take control of their computers from their phones to execute or launch tasks and receive real-time progress notifications.
Doubao has announced an update to its “Work Tasks” mode, adding support for remotely controlling a computer from a phone. Once authorization is complete, users can remotely take control of their computer through the Doubao mobile app to continue unfinished desktop tasks or start new ones, while receiving real-time progress notifications. Doubao can also access file context within the local computer environment, process documents, images, code, spreadsheets, and other materials, and perform more complex computer operations. The use cases listed by Doubao include using a phone to operate a computer and access local files, confirming file versions and converting formats on the computer before sending them back to the phone; having the computer generate summaries and charts and send them to the phone when the user temporarily steps away; and linking different computers through a phone so that two computers can work together to retrieve files and send them to the phone.
OpenAI funds 14 independent projects exploring AI policy for the intelligence age
行业动态
OpenAI announced funding for projects by 14 independent organizations exploring new policies to expand economic opportunity and strengthen societal resilience in the intelligence age. The results are scheduled to be published in 2027.
OpenAI announced grants for 14 projects led by independent organizations in the United States, the European Union, Brazil, Singapore, and South Korea to explore new policy approaches for expanding economic opportunity and strengthening social resilience in the Intelligence Age. The grants fulfill a commitment OpenAI made when it previously published its “Industrial Policy for the Intelligence Age” report, and the call for proposals attracted responses from more than 400 individuals and organizations. OpenAI will provide a total of US$1 million in funding and up to US$1 million in API credits. Some projects will produce research and policy models, while others will build prototypes, datasets, and frameworks that can be tested in practice. All projects will run for six months, with results to be released in 2027.
Higgsfield raises $400 million in Series B funding at a $5.4 billion valuation
行业动态
AI video and image creation platform Higgsfield announced that it has raised $400 million in a Series B round led by DST Global, valuing the company at $5.4 billion.
Higgsfield announced that it has completed a US$400 million Series B funding round at a valuation of US$5.4 billion, more than four times its US$1.3 billion valuation in the previous round. The round was led by DST Global, with participation from multiple new and existing investors. The funding will be used for R&D, expanding global infrastructure, recruiting AI talent, and scaling up global marketing.
Groq raises $350 million in Series A funding at a $3.5 billion valuation
行业动态
Groq announced that it has raised $350 million in a Series A round led by Disruptive, with NVIDIA planning to participate, valuing the company at $3.5 billion.
Groq has completed a US$350 million Series A funding round led by Disruptive, with NVIDIA planning to participate, at a valuation of US$3.5 billion. The company is transitioning from an AI chip manufacturer into a neocloud provider and currently operates 13 data centers serving more than 6 million developers. Groq plans to expand its computing capacity from 54 megawatts to more than 200 megawatts by 2027, with the funding supporting demand for access to medium- and large-scale NVIDIA accelerated computing clusters.
OpenAI urges organizations to strengthen AI security
技术与洞察
OpenAI published a post outlining its security response following the Hugging Face breach and proposed four pillars of defense. It also urged organizations to immediately use AI to strengthen their security teams and take advantage of the defensive window that is already open.
OpenAI published a post outlining its security defense strategy following the intrusions involving OpenAI and Hugging Face, along with an action checklist for organizations. OpenAI’s defense framework rests on four pillars: using Codex and security plugins to validate code before deployment and help remediate vulnerabilities; having AI triage nearly all initial security alerts before escalating them to human analysts; continuously using frontier intelligence to enumerate and block potential attack paths; and investing heavily in foundational security measures such as defense in depth and least privilege. Regarding access, OpenAI has made its cybersecurity capabilities available only to trusted defenders since earlier this year. The post recommends that security teams apply for Trusted Access for Cyber and use GPT-Daybreak-Blue for authorized defensive work. OpenAI also noted that the latest open-weight model with networking capabilities appears to be scheduled for release by the end of August and is likely to significantly accelerate the evolution of the threat landscape.
ByteDance reportedly signs AI model copyright protection agreement with the Motion Picture Association
前瞻与传闻
According to Reuters, ByteDance has signed a copyright agreement with the Motion Picture Association covering its Seedance and Seedream AI generation models. The association had previously sent ByteDance a cease-and-desist letter over copyright concerns.
ByteDance has signed an agreement with the Motion Picture Association to strengthen copyright protections for its AI video and image generation models, Reuters reported. The agreement covers the Seedance video generation model and the Seedream image generation model. In the preceding months, the Hollywood trade association had questioned ByteDance's handling of intellectual property and sent cease-and-desist letters regarding the tools. Disney and other studios had also raised concerns that these AI tools could generate unauthorized content featuring copyrighted characters and the likenesses of celebrities. ByteDance said newer versions of the models include stronger intellectual property protections. The models are available through services including TikTok, CapCut, and Dreamina, and the two parties said they would continue working together on safeguards for copyrighted content.
Note: This content was created with AI assistance and may contain hallucinations and errors.
