Tibo says the Codex team promised daily improvements for 28 days or a full reset
要闻
OpenAI Codex lead Tibo outlined a daily delivery commitment for the next 28 days: if the team does not ship a clear improvement with practical value for most Codex and Work users on a given day, it will fully reset usage limits once.
OpenAI Codex lead Tibo said that, during the next 28 days he described, the team would either ship one clear improvement with practical value for most Codex and Work users each day or fully reset usage limits once. The commitment presents daily product improvements and usage resets as the two alternatives. He had previously said that the team had settled on its direction and would work only on product simplification, usage efficiency, new features, and new models. According to his account, user feedback explicitly calls for a simpler product.
崔添翼: “Everything is a plugin” is the founding vision of DeepSeek Harness
开发生态
DeepSeek Harness released v0.2.1-alpha.1 with an experimental Claude Code Mods compatibility layer. According to the official release notes, this stage tests the relationship between the API and plugin capabilities rather than delivering full compatibility; its author, Cui Tianyi, said not all Mods can currently run seamlessly.
DeepSeek Harness added an experimental Claude Code Mods compatibility layer in its v0.2.1-alpha.1 prerelease; the source does not specify a release date. According to the official release notes, this stage primarily tests whether Claude Code Mods API functionality is broadly a subset of DeepSeek Harness plugin capabilities, rather than providing users with full compatibility in practice. Cui Tianyi, the layer’s author and a team member, explained on Zhihu that the feature still has many shortcomings and cannot guarantee one hundred percent compatibility. He said achieving that goal could be technically possible in the future if it warrants the investment. He also stated that using plugins to provide capabilities was a principle established at the project’s inception, not something adopted under pressure or copied from others. The team aims to meet diverse user needs through plugins and the community, address enterprise deployment requirements across industries through a plugin-based approach, and advance research into Agent self-evolution. Creation mode already allows an Agent to write and install plugins, and the team is recruiting researchers in the field.
Musk confirmed on X that SpaceXAI will be renamed SpaceXSI, but provided no specific timeline; the names on its X account and website have not yet been updated. He subsequently posted that SpaceX is a superintelligence company.
Musk confirmed that SpaceXAI will be renamed SpaceXSI while replying to a user’s question on X, but did not say when the change would take effect. As of the source article’s publication, the names on SpaceXAI’s X account and website had not been updated. The reply addressed a direct question about whether the name would change, and Musk subsequently made a separate post describing SpaceX as a superintelligence company. Earlier, Trump had signed an executive order requiring US executive departments to use Super Intelligence and SI instead of Artificial Intelligence and AI in documents such as official communications and public statements.
Reflection reportedly set to release its first open-weight model
前瞻与传闻
Reflection, which received an $800 million investment from Nvidia, is preparing its first open-weight model, according to Axios, with no release date or performance details announced. The company also plans to enable enterprises or governments to combine their own data and compute to build customized AI systems.
AI startup Reflection is preparing to release its first open-weight model soon, according to Axios, but the specific release date and performance details remain undisclosed, and the company declined to comment. Reflection has received an $800 million investment from Nvidia. The report expects its first model to initially lag behind the most advanced U.S. models while competing with China's leading open-weight models. Beyond the model release, the company plans to pursue AI factories, where enterprises or governments would combine their own data, Reflection models and their own compute to build low-cost, customized AI systems. In March 2026, Reflection signed a memorandum of understanding with South Korea's Shinsegae Group to build a 250-megawatt AI factory in South Korea that would run Reflection models and use Nvidia chips.
