Viral AI project J-Space faces fraud allegations, with community retest results "completely opposite" to official data.

According to MaxForAI, the AI project J-Space Cognition Suite, which went viral on X today, has faced community skepticism over its claimed DeepSeek V4 test results, which critics say cannot be reproduced. The project previously asserted that V4 Flash paired with J-Space could match GLM-5.3, while V4 Pro outperforms Fable 5 on multiple agent benchmarks, delivering a 2.53x speed boost and 2.21x improvement in token efficiency. GitHub user GoForceX ran a high-concurrency retest using the 87-question subset of Terminal Bench 2.1, confirming J-Space modules were loaded. The results showed that adding J-Space actually led to a slight drop in benchmark scores, alongside higher token usage and costs — the opposite of the project’s stated gains in performance, speed, and token efficiency. The community has since demanded the project disclose full evaluation configurations, per-question results, operation logs, original time and token consumption data. To date, the project has only published aggregated results and has not provided complete original experimental records to verify these metrics. Notably, the project’s author previously admitted the related data was “indeed exaggerated,” noting actual improvements are roughly between 1.6x and 3x. As of press time, the author has not issued an official response to the community’s queries, and some related skeptical GitHub Issues have been deleted.