Bindu Reddy predicted that open-source AI models have a window of just eight to 12 weeks to improve dramatically. In her view, they must enter what she called a “self-improving loop” or fall behind frontier AI labs.
Reddy also claimed frontier labs are already in that loop phase and said she expected major open-source announcements soon. She did not define what would qualify as a self-improving loop, and the comparison was presented as her assessment.
Anticipation without confirmed dates
Later the same day, Reddy said multiple US open-source models were coming soon and relayed a claim about stealth foundation-model startups preparing open-weight releases. She also described herself as “skeptical but hopeful.”
One commentator said they would be surprised if Reflection could train a model matching GLM 5.3 without years of experience, though they allowed that it might succeed.
A separate rumor watchlist discussed possible timing for Qwen, Kimi and GLM models while labeling the Kimi timing speculative and the GLM window an estimate. The same post said there was no credible release window for DeepSeek V4.2.
These posts show anticipation among the people sharing them, but they do not confirm a release date, a specific capability jump or Reddy's eight-to-12-week deadline.
